Best Free AI / ML Tools

Vector database. The Starter plan is free: up to 2 GB storage, 2M write units and 1M read units a month, 1 GB of egress a month (reads that return data stop at the cap until the next billing period), 5 indexes, 1 project and 2 users, on AWS us-east-1 only. Assistant includes 1 GB storage and 500k input, 300k output and 500k context tokens a month. Three embedding models include 5M tokens a month each; bge-reranker-v2-m3 includes 500 requests a month. Builder is $20/month flat.

Labelled vector_store. We read that on 2026-09-01 from https://www.pinecone.io/, where it says: “Pinecone is a fully managed vector database built for AI. Writes are instantly searchable, indexing is automatic, and queries stay fast at any scale.”. How we use this

Vector database. The Qdrant Cloud Free Tier is a single-node cluster with 0.5 vCPU, 1 GB RAM and 4 GB disk, for testing and prototypes, with no credit card required; Qdrant says it serves about 1M vectors of 768 dimensions. It includes free Cloud Inference with selected models. An unused free cluster is suspended after 1 week and deleted after 4 weeks of inactivity if not reactivated. Paid clusters are billed on CPU, memory and disk usage.

• An unused Free Tier cluster is suspended after 1 week and deleted after 4 weeks of inactivity unless you reactivate it. (From qdrant.tech/documentation/cloud/create-cluster, read 2026-10-07.)

Labelled vector_store. We read that on 2026-09-01 from https://qdrant.tech/pricing/, where it says: “Scale to production with fully managed vector search”. How we use this

LLM observability and evaluation platform by LangChain — Developer plan is $0/seat/month with up to 5k base traces/month, then pay-as-you-go. LCU is $1.50/unit and LSU is $1.00/unit. Plus plan is $39/seat/month with up to 10k base traces/month, then pay-as-you-go.

Labelled llm_observability. We read that on 2026-09-01 from https://www.langchain.com/pricing, where it says: “Observability See exactly what your agents are doing”. How we use this

Rebranded to Respan; keywordsai.co redirects to respan.ai. Free plan: full platform, 100k logs, 1k scores, 5 datasets, 2 evaluators, 5 prompts. No credit card required. We could not read the page we cite for this offer when we last looked, on 2026-10-31, so we cannot confirm these terms today.

Labelled llm_observability. We read that on 2026-09-02 from https://www.respan.ai, where it says: “Observe calls in traces, watch spend in metrics, and run evals”. How we use this

Comet makes Opik, for LLM and agent observability and evaluation, and an MLOps platform for ML experiment tracking. Opik is open source and free to self-host. Opik Free Cloud allows up to 10 team members, 25k spans per month and 60-day data retention, with agent tracing and analysis, test suites and assertions, and Agent Playground. MLOps Free covers 1 platform user, with experiment tracking, dataset management and versioning, Model Registry and 100GB of data storage. Academics can apply for a free Pro plan.

Labelled llm_observability. We read that on 2026-09-01 from https://www.comet.com/site/pricing/, where it says: “Full AI observability & agent testing feature set”. How we use this

Open-source LLM engineering platform for tracing, evaluating and debugging AI applications. The Langfuse Cloud Hobby plan is free with no credit card: 50k units a month (each trace, observation and score counts as one unit), 30 days of data access, 2 users, community support via GitHub, and all platform features with limits. Core is $29/month with 100k units, then $8 per 100k units up to 1M units and less above that. The open-source self-hosted edition is free under the MIT license.

Labelled llm_observability. We read that on 2026-09-01 from https://langfuse.com/, where it says: “Trace, evaluate, and improve AI agents with one open platform”. How we use this

Platform for agent observability, evaluation and improvement. AX Free costs $0 with no credit card required: 25k trace spans per month, 1 GB ingestion per month, 15 days of retention, unlimited users and evals, 10 Signal issues per month, the Alyx agent and community support. AX Pro is $50 per month with 50k spans and 10 GB of ingestion per month and 30 days of retention. Arize also offers Phoenix, which is ELv2 licensed and can be self-hosted, and 2 Phoenix Cloud instances for free. When we last read the page we cite for this offer, on 2026-11-08, we refused the change we considered recording because it named no figure that had moved, and refusing a change is not a confirmation of the terms above, so we cannot confirm these terms today.

Labelled llm_observability. We read that on 2026-09-01 from https://arize.com/pricing, where it says: “Arize AX is the AI native platform for agent observability”. How we use this

Evals, prompt playground, and data management for Gen AI. Starter plan (free): 1 GB processed data/month, 10k scores/month, 14 days data retention. Unlimited users, projects, datasets, playgrounds, and experiments. Overage: $4/GB data, $2.50 per 1k scores.

Labelled llm_observability. We read that on 2026-09-01 from https://www.braintrust.dev/pricing, where it says: “Free usage is included for traces, evals, and your whole team”. How we use this

ML platform (Gradient Notebooks, Workflows and Deployments) from Paperspace, now part of DigitalOcean. The Free plan is $0 with public projects, 5 GB of storage and free machines for Notebooks in your private workspace (Free CPU C4 and Free GPU M4000; free machines shut down after at most 6 hours). Other machines are billed per hour. Pro is $8/month for individuals and $12/month for teams; Growth is $39/month. DigitalOcean says new customers can still sign up but recommends its own GPU products instead.

Labelled model_hosting. We read that on 2026-09-01 from https://www.paperspace.com/pricing, where it says: “Deployments (inference) - Model Repository”. How we use this

Computer vision platform for labeling, training and deploying models. On September 18, 2026 Roboflow announced a free tier for its Core plan: 10 credits that refresh every month, private projects and models, and model weight download. More credits can be bought. Credits pay for storage, labeling, training and cloud deployment; running models on your own hardware with Roboflow Inference uses none. Core's data licensing and model training setting is on by default, including for private projects; a workspace can turn it off. Public plan workspaces are moving to the Core free tier. The page we cite for Roboflow states in words that it is free, which confirms the price — the limits stated here come from our own record rather than from that page. Note — 2026-07-30: Roboflow's free Public plan went from "$60/mo free credits" to "15 credits / month" between the Internet Archive's 2026-06-23 and 2026-07-30 captures of its pricing page. Roboflow states no dollar value for a credit; its Core plan sells 50 credits for $99 a month. Source ↗

Labelled model_hosting. We read that on 2026-09-01 from https://roboflow.com/pricing, where it says: “Data labeling suite w/ AI features Model training Workflow builder Cloud hosted deployment”. How we use this

Serverless platform for running CPU and GPU workloads. The Starter plan has no monthly fee and includes $30 of compute per month (GPU, CPU and memory), 3 workspace seats, 100 containers and 10 GPU concurrency. Usage beyond the $30 is billed per second: T4 $0.000164 per second; H100 $0.001097 per second. Since 2026-09-01, Shared Endpoints are billed per token from the first request and are not covered by the $30. Since 2026-10-01, network egress beyond 1 TiB a month is billed at $0.04 per GiB. Team is $250 per month with $100 of compute included. Note — 2026-10-01: Since October 1, 2026, Modal charges for network egress: the Starter plan includes 1 TiB a month, Team 10 TiB and Enterprise 100 TiB, and egress beyond that costs $0.04 per GiB. Source ↗

• Modal's billing docs say a payment method must be on file to use Modal. (From modal.com/docs/guide/billing, read 2026-10-23.)
• Modal's terms say Modal will not train any AI model on customer data, or send customer data into large language models, without the customer's prior written consent. (From modal.com/legal/terms, read 2026-10-23.)

Labelled model_hosting. We read that on 2026-10-31 from https://modal.com/pricing, where it says: “Serve LLMs, image, video, and audio models, from real-time to batch”. How we use this

ML model hub. Since 2026-10-07, free accounts get no included Inference Providers credits: free users buy credits or upgrade to PRO to call Inference Providers, which serves 200+ models. PRO ($9/month) gets $2.00 a month. Free accounts get 100GB of private storage and best-effort public storage; substantial storage needs PRO, Team or Enterprise. Note — 2026-10-07: Hugging Face stopped giving free accounts included monthly Inference Providers credits; free users now have to buy credits or upgrade to PRO to call Inference Providers. Source ↗

• A free Hugging Face account needs purchased credits, which take a payment method, to call Inference Providers. (From huggingface.co/docs/inference-providers/pricing, read 2026-10-22.)

Labelled model_hosting. We read that on 2026-09-01 from https://huggingface.co/docs/inference-providers/pricing, where it says: “Inference Endpoints (dedicated)”. How we use this

LLM observability and evaluation platform by LangChain — Developer plan is $0/seat/month with up to 5k base traces/month, then pay-as-you-go. LCU is $1.50/unit and LSU is $1.00/unit. Plus plan is $39/seat/month with up to 10k base traces/month, then pay-as-you-go.

Labelled llm_evaluation. We read that on 2026-09-01 from https://www.langchain.com/pricing, where it says: “Evaluation Score and improve agent performance”. How we use this

Rebranded to Respan; keywordsai.co redirects to respan.ai. Free plan: full platform, 100k logs, 1k scores, 5 datasets, 2 evaluators, 5 prompts. No credit card required. We could not read the page we cite for this offer when we last looked, on 2026-10-31, so we cannot confirm these terms today.

Labelled llm_evaluation. We read that on 2026-09-02 from https://www.respan.ai, where it says: “Route, observe, and evaluate every LLM call”. How we use this

Comet makes Opik, for LLM and agent observability and evaluation, and an MLOps platform for ML experiment tracking. Opik is open source and free to self-host. Opik Free Cloud allows up to 10 team members, 25k spans per month and 60-day data retention, with agent tracing and analysis, test suites and assertions, and Agent Playground. MLOps Free covers 1 platform user, with experiment tracking, dataset management and versioning, Model Registry and 100GB of data storage. Academics can apply for a free Pro plan.

Labelled llm_evaluation. We read that on 2026-09-01 from https://www.comet.com/site/pricing/, where it says: “Systematically test your AI application over an entire dataset”. How we use this

Open-source LLM engineering platform for tracing, evaluating and debugging AI applications. The Langfuse Cloud Hobby plan is free with no credit card: 50k units a month (each trace, observation and score counts as one unit), 30 days of data access, 2 users, community support via GitHub, and all platform features with limits. Core is $29/month with 100k units, then $8 per 100k units up to 1M units and less above that. The open-source self-hosted edition is free under the MIT license.

Labelled llm_evaluation. We read that on 2026-09-01 from https://langfuse.com/, where it says: “Evaluation LLM-as-a-judge, heuristic functions, or human review. Run evaluators on production data or during experiments.”. How we use this

Platform for agent observability, evaluation and improvement. AX Free costs $0 with no credit card required: 25k trace spans per month, 1 GB ingestion per month, 15 days of retention, unlimited users and evals, 10 Signal issues per month, the Alyx agent and community support. AX Pro is $50 per month with 50k spans and 10 GB of ingestion per month and 30 days of retention. Arize also offers Phoenix, which is ELv2 licensed and can be self-hosted, and 2 Phoenix Cloud instances for free. When we last read the page we cite for this offer, on 2026-11-08, we refused the change we considered recording because it named no figure that had moved, and refusing a change is not a confirmation of the terms above, so we cannot confirm these terms today.

Labelled llm_evaluation. We read that on 2026-09-01 from https://arize.com/pricing, where it says: “Trace, evaluate, and improve your agents”. How we use this

Evals, prompt playground, and data management for Gen AI. Starter plan (free): 1 GB processed data/month, 10k scores/month, 14 days data retention. Unlimited users, projects, datasets, playgrounds, and experiments. Overage: $4/GB data, $2.50 per 1k scores.

Labelled llm_evaluation. We read that on 2026-09-01 from https://www.braintrust.dev/pricing, where it says: “Transparent pricing for AI evaluation, monitoring, and observability”. How we use this

Data labeling platform (Catalog, Annotate and Model). The Free plan includes 500 Labelbox Units (LBUs) each month, up to 30 users, 50 projects and 25 ontologies. When the LBUs run out you can still access and export data, but cannot add data rows, labels or predictions until the next billing period. It handles image, video, text, audio, document, chat and geospatial data. The Starter plan costs $0.10 per LBU. Students and faculty at qualifying institutions can apply for a free Education License for non-commercial use.

Labelled data_labeling. We read that on 2026-09-01 from https://labelbox.com/, where it says: “The data, environments, and evaluation infrastructure the world's frontier AI labs build on”. How we use this

Computer vision platform for labeling, training and deploying models. On September 18, 2026 Roboflow announced a free tier for its Core plan: 10 credits that refresh every month, private projects and models, and model weight download. More credits can be bought. Credits pay for storage, labeling, training and cloud deployment; running models on your own hardware with Roboflow Inference uses none. Core's data licensing and model training setting is on by default, including for private projects; a workspace can turn it off. Public plan workspaces are moving to the Core free tier. The page we cite for Roboflow states in words that it is free, which confirms the price — the limits stated here come from our own record rather than from that page. Note — 2026-07-30: Roboflow's free Public plan went from "$60/mo free credits" to "15 credits / month" between the Internet Archive's 2026-06-23 and 2026-07-30 captures of its pricing page. Roboflow states no dollar value for a credit; its Core plan sells 50 credits for $99 a month. Source ↗

Labelled data_labeling. We read that on 2026-09-01 from https://roboflow.com/pricing, where it says: “Label images fast with AI-assisted data annotation”. How we use this

AI media gateway: one OpenAI-compatible API to image and video generation models from several providers, billed at each provider's price with no markup, platform fee or subscription. One model, FLUX.1 [schnell] FP8, is listed free per image ("Free to try", served through Fireworks AI). New accounts also get a $1 credit with no card required; Lumenfall's terms say promotional credits typically expire after 90 days. After that you top up prepaid credits. Under Lumenfall's terms (2026-03-05), content sent through free features, including free API usage, may be used to train AI models.

Labelled image_video_generation. We read that on 2026-09-01 from https://lumenfall.ai/, where it says: “The Unified API for AI Image and Video Generation”. How we use this

Keyless Classic API: one image model and GPT-OSS 20B text, free, no sign-up, rate-limited per IP. Every other model needs an API key and Pollen, and free Pollen comes only from quests. No credit card is required. On 2026-10-11, the nine setup quests paid 3 Pollen in total, worth about $3 because 1 Pollen is worth about $1. The documented API for text, image, audio and video requires a GitHub sign-in and an API key. On the keyed API, free use draws from Quest Pollen, with no promise of future grants. Paid-only models, such as Claude models, DeepSeek V4 Pro and Gemini 4 Argon, need purchased Pollen. Gemini 3.8 Flash (google/gemini-3.8-flash) is paid-only on Pollinations, so Quest Pollen cannot pay for it. Kimi K3 (moonshotai/kimi-k3) is not paid-only. Kimi K3.1 Pro (moonshotai/kimi-k3.1-pro) was paid-only until 2026-11-09, when Pollinations made it Quest-accessible; Quest Pollen can now pay for it. To run Claude Code on free Quest Pollen, run "polli harness claude-code on", start Claude Code with "claude --settings ~/.pollinations/claude-code.json", and use a model that is not paid-only. Quest Pollen pays for any model that Pollinations' live catalog (enter.pollinations.ai/models) does not mark paid-only. On 2026-10-18, more than 30 text models were not paid-only, among them GPT-6 Sol, GPT-6.1 Sol, GPT-6 Luna, GLM 5.3, Kimi K3, DeepSeek V4 Flash and DeepSeek V4.1 Flash. The command "polli harness claude-code on" connects Claude Code to Pollinations and uses openai/gpt-6-sol unless you specify another model. Pollinations' harness documentation says the command writes "a Claude Code settings file" and: "Claude Code reads the file only when launched with --settings; plain claude keeps your native login and ~/.claude settings." (From https://github.com/pollinations/pollinations/blob/main/CODING_HARNESSES.md, read 2026-10-20.) GPT-6 Sol costs 1.5 Pollen per million input tokens and 7.5 Pollen per million output tokens. For requests over 272,000 prompt tokens, GPT-6 Sol costs 3 Pollen per million input tokens and 11.25 Pollen per million output tokens, and these rates apply to the whole request. GPT-5 Nano costs 0.0375 and 0.30. GPT-OSS 20B costs 0.05 and 0.18. NVIDIA Nemotron 3.5 Lightning costs 0.045 and 0.165. Adding --model <id> to "polli harness claude-code on" picks another model. The documented API is OpenAI-compatible at the base URL https://gen.pollinations.ai/v1, with an API key from https://enter.pollinations.ai/keys. Pollinations' MCP documentation gives this Claude Code command: "claude mcp add --transport http pollinations https://gen.pollinations.ai/mcp/pollinations --header 'Authorization: Bearer YOUR_KEY'". Its tools include generateImage, generateText, generateVideo, generateAudio, listModels and getBalance. "Calls use the same Pollen wallet as the Pollinations API." The "polli harness claude-code on" command runs Claude Code itself on a Pollinations model; the MCP server gives an agent Pollinations tools. Model IDs of some of the models Quest Pollen can pay for: openai/gpt-6-sol, openai/gpt-6.1-sol, openai/gpt-6-luna, z-ai/glm-5.3, moonshotai/kimi-k3, deepseek/deepseek-v4-flash, deepseek/deepseek-v4.1-flash, openai/gpt-5-nano, openai/gpt-oss-20b, nvidia/nemotron-3.5-lightning, moonshotai/kimi-k3.1-pro. Note — 2026-07-10: Pollinations ended the free Pollen that refilled every hour on every account (0.01 to 0.4 Pollen an hour, by tier). Free Pollen now comes only from Quests, whose rewards are paid once and land as Quest Pollen; further use needs purchased Pollen. The change did not touch the keyless image and text endpoints, which Pollinations documents as the free Classic API. Source ↗

• Pollinations' Classic API needs no sign-up and no key. image.pollinations.ai (one image model; JPEGs up to 768 × 768 with a small Pollinations logo) and text.pollinations.ai (GPT-OSS 20B, with an OpenAI-compatible endpoint at text.pollinations.ai/openai) are free and rate-limited per IP address: one image request at a time, at least 45 seconds apart, and one text request at a time, at least 30 seconds apart. Past the limit they return HTTP 402 with a Retry-After header. Pollinations runs it on a best-effort basis with no uptime guarantee, and says limits and models can change. Every other model needs a key from enter.pollinations.ai. (From github.com/pollinations/pollinations/blob/master/APIDOCS.md, read 2026-11-06.)
• Pollinations' API docs publish no fixed request rate for secret keys, which carry their own model permissions and budgets. A request can get HTTP 429 with a Retry-After header saying how long to wait. (From github.com/pollinations/pollinations/blob/main/APIDOCS.md, read 2026-10-23.)
• Pollinations' Terms of Service state that model licences vary and commercial use depends on each model's licence. The Terms set no separate rules for usage paid with free Quest Pollen. There is no SLA. (From pollinations.ai/terms, read 2026-10-30.)
• Pollinations does not use prompts or outputs to train or fine-tune models without explicit opt-in. Prompts and outputs are processed transiently and not retained beyond what is needed to run the service, except saved configuration and short-lived response caches. A request to a community-hosted model is sent to that provider, whose own terms govern storage and training; Pollinations does not verify those practices. (From pollinations.ai/privacy, read 2026-10-30.)
• Pollinations stores generated images, video, audio, and 3D files at media.pollinations.ai. Retrieval is public with the URL. These files have a 30-day lifecycle from upload or refresh. (From github.com/pollinations/pollinations/blob/main/APIDOCS.md, read 2026-11-03.)
• Pollinations requires browser apps to use a user-authorized sk_ key from Pollinations Pay. Personal sk_ keys must not be in browser code. A pk_ key only identifies the app and returns 401 if sent as a key. (From github.com/pollinations/pollinations/blob/main/APIDOCS.md, read 2026-11-03.)
• Pollinations Pay, also called BYOP, lets users authorize an app to spend Pollen from their own wallet. Users set budget, expiry, models, and permissions. They can edit or revoke issued keys. (From github.com/pollinations/pollinations/blob/main/enter.pollinations.ai/POLLEN_FAQ.md, read 2026-11-03.)
• Pollinations' image endpoint has a safe parameter. It defaults to off. The value nsfw enables sexual and violence filtering. The value true enables privacy and secrets filtering. (From github.com/pollinations/pollinations/blob/main/APIDOCS.md, read 2026-11-03.)
• Owners of public community models in Pollinations can set prices. Pollinations credits owners 75% of the Pollen spent on their models. A linked GitHub username is required. (From github.com/pollinations/pollinations/blob/main/enter.pollinations.ai/POLLEN_FAQ.md, read 2026-11-03.)

Labelled image_video_generation. We read that on 2026-09-01 from https://pollinations.ai/, where it says: “Generate images, text, audio and video with state-of-the-art AI models”. How we use this

Workers AI runs AI models on Cloudflare's global network. Every account gets 10,000 Neurons per day at no charge, reset daily at 00:00 UTC; on the Workers Free plan, requests beyond that fail until the reset. On the Workers Paid plan (from $5 per month), usage above 10,000 Neurons per day costs $0.011 per 1,000 Neurons. Eight models require the Workers Paid plan or prepaid AI Gateway credits: @cf/moonshotai/kimi-k2.6, @cf/moonshotai/kimi-k2.7-code, @cf/zai-org/glm-5.2, @cf/zai-org/glm-5.3, @cf/zai-org/glm-5.3-flash, @cf/deepseek-ai/deepseek-v4-flash-0731, @cf/deepseek-ai/deepseek-v4.1-flash-1017 and @cf/deepseek-ai/deepseek-v4-pro-0813. Cloudflare removes 24 models from Workers AI on 2026-11-20, including six of those eight (all but glm-5.3 and deepseek-v4.1-flash-1017); they can be called until then, and from 2026-11-20 requests to them fail with HTTP 410 Gone (error 5028). On 2026-05-30, Cloudflare deprecated 18 models and told users to remove references to them: @cf/meta/llama-3.1-8b-instruct, @cf/meta/llama-3.1-8b-instruct-awq, @cf/meta/llama-3.1-70b-instruct, @cf/meta/llama-3-8b-instruct, @cf/meta/llama-3-8b-instruct-awq, @hf/meta-llama/meta-llama-3-8b-instruct, @cf/meta/llama-2-7b-chat-int8, @cf/meta/llama-2-7b-chat-fp16, @cf/mistral/mistral-7b-instruct-v0.1, @hf/mistral/mistral-7b-instruct-v0.2, @hf/google/gemma-7b-it, @cf/google/gemma-3-12b-it, @hf/nousresearch/hermes-2-pro-mistral-7b, @cf/microsoft/phi-2, @cf/defog/sqlcoder-7b-2, @cf/unum/uform-gen2-qwen-500m, @cf/facebook/bart-large-cnn and @cf/moonshotai/kimi-k2.5. On 2026-10-15, Cloudflare's model catalog still listed @cf/meta/llama-3.1-8b-instruct-fp8, @cf/meta/llama-3.3-70b-instruct-fp8-fast and the -lora variants, but no longer listed @cf/meta/llama-3.1-8b-instruct-fast; llama-3.1-8b-instruct-fp8 and the -lora variants are among the 24 models removed on 2026-11-20. Cloudflare's suggested replacements were @cf/zai-org/glm-4.7-flash, @cf/google/gemma-4-26b-a4b-it and @cf/moonshotai/kimi-k2.6, which is itself removed on 2026-11-20. Since 2026-05-30, requests to @cf/moonshotai/kimi-k2.5 go to @cf/moonshotai/kimi-k2.6, which costs more and requires the Workers Paid plan or prepaid AI Gateway credits; without them, requests get HTTP 403 (error 5035). Note — 2026-07-28: On 2026-07-28 Cloudflare took Kimi K2.6, Kimi K2.7 Code and GLM-5.2 off the Workers Free plan; they now need the Workers Paid plan (or, from 2026-08-07, prepaid AI Gateway credits). Source ↗

• Workers AI's free allowance on the Workers Free plan needs no credit card. (From cloudflare.com/products/workers-ai, read 2026-10-22.)
• Cloudflare prices models in Neurons per million tokens. @cf/openai/gpt-oss-20b costs 18,182 Neurons per million input tokens and 27,273 per million output tokens. @cf/openai/gpt-oss-120b costs 31,818 and 68,182 Neurons respectively. @cf/meta/llama-3.3-70b-instruct-fp8-fast costs 26,668 and 204,805 Neurons. @cf/meta/llama-3.1-8b-instruct-fp8-fast costs 4,119 and 34,868 Neurons. 10,000 Neurons pay for about 367,000 output tokens of gpt-oss-20b, 147,000 of gpt-oss-120b, 49,000 of llama-3.3-70b-instruct-fp8-fast, or 287,000 of llama-3.1-8b-instruct-fp8-fast, not counting input tokens. Other models have their own rates on Cloudflare's pricing page. (From developers.cloudflare.com/workers-ai/platform/pricing, read 2026-11-12.)
• Workers AI is subject to Cloudflare's Self-Serve Subscription Agreement. The agreement does not restrict commercial use. (From developers.cloudflare.com/workers-ai/platform/data-usage, read 2026-10-29.)
• Cloudflare does not create or train the third-party AI models in Workers AI. Each model may be subject to license terms from its provider. (From developers.cloudflare.com/workers-ai/platform/data-usage, read 2026-10-29.)
• Cloudflare does not use Workers AI inputs or outputs to train models. It does not retain them after a standard inference request, unless you enable a feature that stores them, such as logging or caching in AI Gateway. (From developers.cloudflare.com/workers-ai/platform/data-usage, read 2026-10-29.)
• Cloudflare AI Search is a managed search service over your own content. Every account gets 5 million ingestion tokens, 10 GB-month of storage, 1,000 semantic queries and 1,000 full-text queries a month. Embedding and reranking calls are billed as AI Search usage and do not appear on the Workers AI bill. (From developers.cloudflare.com/ai-search/platform/limits-pricing, read 2026-11-03.)

Labelled llm_api. We read that on 2026-09-01 from https://developers.cloudflare.com/workers-ai/platform/pricing/, where it says: “prepaid AI Gateway credits to pay for Workers AI inference”. How we use this

AI model API for Command chat models, Embed, Rerank, Transcribe and Parse. Every account starts with a Trial API key: calls made with it are free and limited to 1,000 API calls a month. Trial rate limits: 20 requests a minute per Chat model, 2,000 Embed inputs a minute and 10 Rerank requests a minute. Trial keys can use all of Cohere's models and APIs. Production keys are pay-as-you-go. Prices per 1M tokens: Command R7B $0.0375/$0.15; Command R $0.15/$0.60. The pricing page lists Command A+ (Apache 2.0) at $0 through an API key and as a model download.

• Cohere's Trial API keys may not be used for production or commercial purposes. (From cohere.com/pricing, read 2026-10-04.)
• Cohere creates a free Trial API key with each new account; no credit card is needed. (From cohere.com/pricing, read 2026-10-22.)

Labelled llm_api. We read that on 2026-09-01 from https://cohere.com/pricing, where it says: “Command Generative language models”. How we use this

Hetzner's experimental Inference API: an OpenAI-compatible API for open-weight models, free while it stays experimental. Base URL: https://inference.hetzner.com/api/v1. Models listed on 2026-10-15, with the IDs Hetzner's docs give: Qwen/Qwen3.6-35B-A3B-FP8 and Qwen3.8-27B, each with a 262,144-token context and image input. Hetzner says the list will change while the API is experimental; GET /v1/models returns the current one. Limits per API key, per minute: 4M input tokens, 100k output tokens and 10 requests. Needs a Hetzner account; keys are created at experiments.hetzner.com. Hetzner says it does not store the content of requests or responses. The page we cite for Hetzner Inference API states in words that it is free, which confirms the price — the limits stated here come from our own record rather than from that page.

• Hetzner says not to use the Inference API for production: it is experimental and offered as is, with no guarantee of performance or availability. (From docs.hetzner.com/general/company-and-policy/experiments/inference, read 2026-10-09.)
• It is free only while it stays experimental. Hetzner says it will email users before that changes. (From docs.hetzner.com/general/company-and-policy/experiments/inference, read 2026-10-09.)

Labelled llm_api. We read that on 2026-10-09 from https://docs.hetzner.com/general/company-and-policy/experiments/inference/, where it says: “The Inference API provides access to open-source large language models via an OpenAI-compatible REST API.”. How we use this

Free tier: Gemini 4 Hydrogen, Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, Gemini 3 Flash Preview, Gemini Embedding 2 and Gemma 4 are free of charge. Also free of charge are Gemini 3.8 Live (real-time voice), Gemini 3.5 Live Translate (translation), Gemini 3.5 Transcribe (speech-to-text), Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS (text-to-speech). Gemini Decision (gemini-decision-1), released 2026-11-11, is a multimodal decision model available through the Decisions API. It answers yes/no, choice and rubric questions with probabilities and generates zero output tokens, so it has no output price; input is free of charge on the free tier with 10,000 requests per day, and $0.10 per million input tokens on the paid tier. Gemini 4 Hydrogen (gemini-4-hydrogen), released 2026-11-12, is free of charge on the Gemini API free tier for input, output and context caching. On the paid tier, Gemini 4 Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens (output includes thinking tokens), with context caching at $0.075 (implicit caching only); on the Batch API, Gemini 4 Hydrogen is $0.75 input and $3.75 output. The other Gemini 4 models, Gemini 4 Carbon (gemini-4-carbon) and Gemini 4 Argon (gemini-4-argon), are paid only, as are Gemini 3.1 Pro Preview and Gemini Omni 1.2 Flash (gemini-omni-1.2-flash). Gemini 4 Carbon, which Google calls its most capable reasoning model, has been generally available since 2026-11-10 and is not available on the Gemini API free tier. Gemini 4 Argon, generally available since 2026-10-13, is not available on the Gemini API free tier. According to ai.google.dev/gemini-api/docs/google-ai-plans, "Gemini 4 Argon is not available on the Free plan." Gemini 4 Argon is available through the Gemini API on the paid tier. "Direct use of the Gemini API (such as using API keys or external applications) is billed and managed separately," per ai.google.dev/gemini-api/docs/google-ai-plans. Google AI Pro and Ultra provide access to Gemini 4 Argon in Google AI Studio, per ai.google.dev/gemini-api/docs/google-ai-plans. Gemini 4 Argon is in Antigravity on Google AI Ultra and paid Google AI Pro. In Google AI Studio, Gemini 4 Carbon requires a Google AI Ultra plan and is not available on the Free or Google AI Pro plans (aistudio.google.com/docs/google-ai-plans). In Antigravity, Gemini 4 Carbon is available on Google AI Ultra and Enterprise only (antigravity.google/docs/models). On a Google AI Pro trial, Argon is only available "on trial with eligible premium devices" (antigravity.google/docs/models). Claude Opus 5.5 and Claude Sonnet 5.5 are not included on Google AI Pro trials; trial users and those with memberships through third-party partners, Pixel Pass or Google Fi manage subscriptions through Google One. Google AI Pro via partner offers like Jio's does not include Gemini 4 Argon, Claude Sonnet 5.5 or Claude Opus 5.5 in Antigravity. Per a Google staff member on discuss.ai.google.dev (2026-10-21), "Those frontier models are restricted to directly billed (non-trial/non-promotional) Google AI Pro, Google AI Ultra and Enterprise plans." No Gemini image model has a free tier in the API. Nano Banana 2.1 (gemini-nano-banana-2.1) became generally available on 2026-10-06. It costs $0.0336 per 1K image, $0.0504 per 2K image and $0.113 per 4K image. The Batch API halves those prices. Nano Banana Pro costs $0.134 per 1K or 2K image and $0.24 per 4K image. Except for Gemini Decision, Google publishes no free-tier rate limits; each project's limits are shown in Google AI Studio. Developers report 429 errors showing 20 requests per day for Gemini 3.x Flash models and 500 requests per day for 3.x Flash-Lite models as of September 2026. Rate limits apply per project, not per API key. Daily request quotas reset at midnight Pacific time, per Google's rate-limits page. Since 2026-09-18 Google serves the Gemini 2.5 models only to users who have used them before, and points new projects to 3.5 Flash-Lite or 3.8 Flash. If your app calls gemini-3.7-flash, Google deprecated it on 2026-10-08. Requests to gemini-3.7-flash are routed automatically to gemini-3.8-flash. Google says to update the model string to gemini-3.8-flash. Google deprecated gemini-2.5-pro and gemini-2.5-flash on 2026-11-02. Both models shut down on 2026-12-01. Use them until that date. From that date, requests to gemini-2.5-pro route to gemini-3.8-flash and are billed at its rates. Requests to gemini-2.5-flash route to gemini-3.5-flash-lite and are billed at its rates. Update your model string before 2026-12-01 and review Google’s migration guides for breaking API changes. gemini-2.5-flash-lite is not deprecated. If your app calls Gemini 3.5 Flash (gemini-3.5-flash), Google deprecated it on 2026-10-08. Requests to gemini-3.5-flash are routed automatically to gemini-3.6-flash. Google says to update the model string to gemini-3.6-flash or gemini-3.8-flash. Google shut down Gemini Omni Flash Preview (gemini-omni-flash-preview) on 2026-10-22. Google's replacement is gemini-omni-1.2-flash. Google shuts down Gemini Omni 1.1 Flash (gemini-omni-1.1-flash) on December 3, 2026; after that, requests that name gemini-omni-1.1-flash are answered by gemini-omni-1.2-flash at Gemini Omni 1.2 Flash rates. Gemini 3.1 Flash-Lite (gemini-3.1-flash-lite) can be called until 2027-05-07, when Google shuts it down. Google's replacement is gemini-3.5-flash-lite. Paid, per million tokens (input/output; output prices include thinking tokens): Gemini 4 Carbon $2/$10 through 2026-11-30 (context caching $0.10, then $0.20, implicit caching only); Gemini 4 Argon $2/$10 through 2026-11-30 (context caching $0.10, then $0.20, implicit caching only); Gemini 3.8 Flash $0.75/$3.75 until 2026-12-31, then $1.50 input and $7.50 output from 2027-01-01 (3.6 Flash: the same prices and dates); Gemini 3.5 Flash-Lite $0.30/$2.50; Gemini 3.1 Flash-Lite $0.25/$1.50; Gemini 3.1 Pro Preview $2/$12 (prompts up to 200K tokens; prompts over 200K tokens cost $4 input and $18 output). Through 2026-11-30, Gemini 4 Argon's Standard output price is below Gemini 3.1 Pro Preview's for prompts up to 200K tokens, with the same input price; from 2026-12-01 Argon's is higher. From December 1, 2026, Gemini 4 Argon will cost $4 per million input tokens and $20 per million output tokens for Standard, $2 input and $10 output for Batch API, $2 input and $10 output for Flex, and $7.20 input and $36 output for Priority. From December 1, 2026, Gemini 4 Carbon will also cost $4 per million input tokens and $20 per million output tokens for Standard, $2 input and $10 output for Batch API and Flex, and $7.20 input and $36 output for Priority. The Batch API costs half the standard price; Gemini 3.8 Flash is $0.375 input and $1.875 output per million tokens through 2026-12-31. Gemini 4 Carbon and Gemini 4 Argon on the Batch API are $1 input and $5 output through 2026-11-30. Google also sells two tiers, Flex and Priority, both in Preview. A request picks one with the service_tier field ("flex" or "priority"); without it the request is Standard. Gemini 4 Carbon and Gemini 4 Argon on Flex are $1 input and $5 output through 2026-11-30; on Priority they are $3.60 input and $18 output through 2026-11-30. "The Gemini Flex API is an inference tier that offers a 50% cost reduction compared to standard rates, in exchange for variable latency and best-effort availability" (ai.google.dev/gemini-api/docs/flex-inference). "Priority inference is priced at 75-100% more than the standard API and billed per token" (ai.google.dev/gemini-api/docs/priority-inference). The Gemini API Contributor Tier is an invite-only private preview. Eligible users will be invited by email. It offers half the Standard price for Gemini 4 Carbon, Gemini 4 Argon, Gemini 3.1 Pro Preview, Gemini 3.8 Flash, and Gemini 3.5 Flash-Lite. Google uses the data from those models to improve its products. "Please don't submit sensitive or personal information" (ai.google.dev/gemini-api/docs/logs-policy). Grounding with Google Search is supported on the free tier for Gemini 3.8 Live, Gemini 3.8 Live Extended Thinking and Gemini 3.1 Flash Live Preview. It is not available on the free tier for Gemini 4 Hydrogen, Gemini 4 Argon or the Gemini 3.x Flash, Flash-Lite and Pro models. On the paid tier, 5,000 search requests a month are free, shared across Gemini 3 and newer models, then $14 per 1,000. Accounts opened after 2026-03-02 cannot spend the $300 Google Cloud welcome credit on the Gemini API.

• The Gemini API Free Tier needs no credit card or billing account; linking a billing account and prepaying moves you to the Paid Tier. (From ai.google.dev/gemini-api/docs/billing, read 2026-10-22.)
• Google's pricing page says free-tier content is used to improve Google's products. (From ai.google.dev/gemini-api/docs/pricing, read 2026-10-04.)
• For developers outside the European Economic Area, Switzerland and the United Kingdom, Google's terms say human reviewers may read free-tier prompts and responses, and Google says not to send sensitive, confidential or personal information on the free tier. (From ai.google.dev/gemini-api/terms, read 2026-10-04.)
• Paid-tier projects created on or after October 26, 2026, store Generate Content API requests and responses by default. Projects created before that date keep their existing Generate Content API storage setting, which is off unless it was turned on in AI Studio. The Interactions API stores requests by default in all projects, including projects created before October 26, 2026. The Paid Tier keeps stored interactions for 55 days and the Free Tier for 1 day. Logging also covers calls made through OpenAI compatibility endpoints. Logs are retained for up to 55 days, configurable to 7, 14 or 28 days. Logs are not used to improve Google's products unless shared. Logging can be toggled independently for each API in AI Studio, and store=false on a request turns it off for that request. (From ai.google.dev/gemini-api/docs/logs-policy, read 2026-11-06.)
• Apps that serve users in the European Economic Area, Switzerland or the United Kingdom may use only the paid tier. (From ai.google.dev/gemini-api/terms, read 2026-10-04.)
• Google's terms require users to be 18 years of age or older to use the Gemini API and Google AI Studio. The API may not be used in a website, app, or other service directed towards or likely to be accessed by individuals under 18. These rules apply to both the free and paid tiers. (From ai.google.dev/gemini-api/terms, read 2026-10-17.)
• Google offers the Gemini API and Google AI Studio, free and paid, in 230 countries and territories, including Taiwan, Brazil, Japan, Singapore, and the United Kingdom. China, Hong Kong, Macao, Russia, Belarus, and Iran are not among them. Google says to developers outside the list: "try the Gemini API in Gemini Enterprise Agent Platform". (From ai.google.dev/gemini-api/docs/available-regions, read 2026-10-17.)
• Google's terms say changes to Gemini API paid prices take effect 30 days after Google posts them unless Google specifies otherwise. Pricing for a new paid service takes effect immediately unless Google specifies otherwise. (From ai.google.dev/gemini-api/terms, read 2026-10-17.)
• Google is moving existing Postpay accounts to Prepay for Gemini API usage. An account that has not switched and added credits by the cutover date in its account notice has its paid service interrupted; other Google Cloud services on the billing account stay on Postpay. (From ai.google.dev/gemini-api/docs/billing, read 2026-10-07.)
• When the Prepay balance reaches $0, every API key in every project on the billing account stops working until credits are added; requests then fail with an HTTP 402 Payment Required error. (From ai.google.dev/gemini-api/docs/billing, read 2026-10-07.)
• The monthly Google Cloud credits from Google AI Pro or Ultra are eligible Google Cloud credits. On the Prepay billing plan, these credits pay for Gemini API usage only after you buy Prepay credits in AI Studio. With a $0 Prepay balance, requests fail even with unused Google Cloud credits, according to a Google staff member on discuss.ai.google.dev. (From discuss.ai.google.dev/t/error-code-402-message-your-prepayment-credits-are-depleted/185524, read 2026-10-10.)
• A paid project does not automatically revert to the free tier, even when its Prepay balance reaches $0. To use the free tier, disable billing on the project, which unlinks it from its billing account. (From ai.google.dev/gemini-api/docs/billing, read 2026-10-10.)
• Disabling billing on a Google Cloud project stops all other paid Google Cloud services in it. Some resources may be deleted and become non-recoverable. Outstanding charges are still due. (From cloud.google.com/billing/docs/how-to/modify-project, read 2026-10-10.)

Labelled llm_api. We read that on 2026-09-01 from https://ai.google.dev/pricing, where it says: “Our first hybrid reasoning model which supports a 1M token context”. How we use this

Free inference endpoints on build.nvidia.com: models marked Free Endpoint, including DeepSeek V4.1 Flash, GLM 5.3 Flash, gpt-oss-20b, Gemma 4 31B and NVIDIA Nemotron 3 Ultra, can be called at no cost. Their model IDs are deepseek-ai/deepseek-v4.1-flash, z-ai/glm-5.3-flash, openai/gpt-oss-20b, google/gemma-4-31b-it and nvidia/nemotron-3-ultra-550b-a55b. Mistral Large 4 (mistralai/mistral-large-4-1t-a52b) is a Free Endpoint on build.nvidia.com as of 2026-10-29. NVIDIA's free embedding models are nvidia/nemotron-3-embed-1b, nvidia/nemotron-3.5-embed-1b and nvidia/nemotron-3.5-embed-8b. The free reranking model is nvidia/nemotron-3.5-rerank-8b. The 3.5 models handle text and document images and were added on 2026-10-20. Up to 40 requests per minute; limits may vary by model, and traffic from other users may cause throttling. NVIDIA's API Trial Terms allow free use for testing and evaluation only, not production. NVIDIA posts an end-of-life date on each hosted model's page on build.nvidia.com; after that date, requests for the model return HTTP 410 Gone. NVIDIA stopped supporting NVIDIA Nemotron 3 Super (nvidia/nemotron-3-super-120b-a12b) after 2026-10-19, and its catalogue no longer lists the model (checked 2026-10-21). Nemotron 3.5 Super VL (nvidia/nemotron-3.5-super-vl-120b-a12b), an early-access model, is a Free Endpoint. NVIDIA retired Kimi K3 (moonshotai/kimi-k3) on 2026-10-15. Other model IDs NVIDIA retired in 2026 include these. DeepSeek: deepseek-ai/deepseek-v4-pro and deepseek-ai/deepseek-v4-flash on 2026-08-07, deepseek-ai/deepseek-v4-pro-0813 on 2026-09-14 and deepseek-ai/deepseek-v4-flash-0731 on 2026-09-21. Kimi: moonshotai/kimi-k2-instruct and moonshotai/kimi-k2-thinking on 2026-05-12. GLM: z-ai/glm4.7 on 2026-05-14, z-ai/glm-5.1 on 2026-07-02, z-ai/glm-5.2 on 2026-08-21 and z-ai/glm-5.3 on 2026-10-27 (GLM 5.3 Flash, z-ai/glm-5.3-flash, is still a Free Endpoint). Qwen: qwen/qwen2.5-coder-32b-instruct on 2026-05-12, qwen/qwen3-next-80b-a3b-thinking on 2026-05-21, qwen/qwen3-coder-480b-a35b-instruct on 2026-06-11, qwen/qwen3.5-122b-a10b on 2026-07-20, and qwen/qwen3.5-397b-a17b and qwen/qwen3-next-80b-a3b-instruct on 2026-07-27. MiniMax: minimaxai/minimax-m2.5 on 2026-05-12, minimaxai/minimax-m2.7 on 2026-07-27 and minimaxai/minimax-m3 on 2026-09-09. gpt-oss: openai/gpt-oss-120b on 2026-09-03. Llama: meta/llama-3.2-11b-vision-instruct on 2026-11-05, meta/llama-4-maverick-17b-128e-instruct on 2026-07-27, and meta/llama-3.3-70b-instruct, meta/llama-3.1-8b-instruct, meta/llama-3.1-70b-instruct, meta/llama-3.2-1b-instruct and meta/llama-3.2-3b-instruct on 2026-08-26. Nemotron: nvidia/llama-3.3-nemotron-super-49b-v1, nvidia/llama-3.3-nemotron-super-49b-v1.5, nvidia/llama-3.1-nemotron-nano-8b-v1, nvidia/llama-3.1-nemotron-nano-vl-8b-v1, nvidia/nvidia-nemotron-nano-9b-v2, nvidia/nemotron-nano-12b-v2-vl and nvidia/nemotron-mini-4b-instruct on 2026-08-26, and nvidia/nemotron-3-nano-30b-a3b on 2026-09-01. Mistral: mistralai/mistral-large-3-675b-instruct-2512 on 2026-07-23, mistralai/mixtral-8x7b-instruct-v0.1 on 2026-07-27 and mistralai/mistral-nemotron on 2026-09-28. Gemma: google/diffusiongemma-26b-a4b-it on 2026-11-12, google/gemma-3-27b-it on 2026-05-12, and google/gemma-3n-e4b-it and google/gemma-3n-e2b-it on 2026-07-27. Phi: microsoft/phi-4-mini-instruct and microsoft/phi-4-multimodal-instruct on 2026-07-15. Others: stepfun-ai/step-3.5-flash, bytedance/seed-oss-36b-instruct, sarvamai/sarvam-m and abacusai/dracarys-llama-3.1-70b-instruct on 2026-07-27. Embedding models: nvidia/llama-3.2-nv-embedqa-1b-v2 and nvidia/llama-3.2-nemoretriever-300m-embed-v1 on 2026-05-18, and nvidia/nv-embedqa-e5-v5, nvidia/nv-embed-v1, nvidia/llama-nemotron-embed-1b-v2 and baai/bge-m3 on 2026-08-25.

• NVIDIA records prompts and outputs from free API use to improve its products and services, including AI models. Do not send confidential information or personal data unless expressly permitted. (From assets.ngc.nvidia.com/products/api-catalog/legal/NVIDIA%20API%20Trial%20Terms%20of%20Service.pdf, read 2026-10-10.)
• Using the API in production, or after the trial credits are used up, needs a separate subscription from NVIDIA or a third-party provider. (From assets.ngc.nvidia.com/products/api-catalog/legal/NVIDIA%20API%20Trial%20Terms%20of%20Service.pdf, read 2026-10-06.)
• Production use of NIM needs an NVIDIA AI Enterprise license. The 1-year list price is $4,500 per GPU. The cloud pay-as-you-go rate is $1 per hour per GPU plus the cloud service provider's instance cost. (From docs.nvidia.com/ai-enterprise/planning-resource/licensing-guide/latest/pricing.html, read 2026-10-29.)
• NVIDIA Developer Program members may run downloadable NIM microservices free for research, application development, and experimentation on up to 16 GPUs. Infrastructure may be cloud, data center, or personal workstation. (From docs.api.nvidia.com/nim/docs/product, read 2026-10-29.)
• The free NVIDIA AI Enterprise trial is 90 days long. It covers production deployments and commercial use. (From docs.api.nvidia.com/nim/docs/product, read 2026-10-29.)

Labelled llm_api. We read that on 2026-09-01 from https://build.nvidia.com, where it says: “Experience the leading models to build enterprise generative AI apps now”. How we use this

Fast LLM inference on Groq's LPU hardware. Free plan: gpt-oss-20b, gpt-oss-120b and Qwen3.8 27B, each at 30 requests a minute, 1,000 requests and 200,000 tokens a day, plus Whisper speech-to-text at 2,000 requests a day. Each of the three free models is also limited to 8,000 tokens a minute, and Groq applies whichever limit is reached first. Whisper is also limited to 20 requests a minute, 7,200 audio seconds an hour and 28,800 audio seconds (8 hours of audio) a day. On the free plan, gpt-oss-safeguard-20b is limited to 3 requests a minute, 1,000 requests a day, 2,000 tokens a minute and 200,000 tokens a day. Qwen3.8 27B, gpt-oss-safeguard-20b, Orpheus Arabic Saudi and Orpheus V1 English are Preview models. Groq says Preview models "are intended for evaluation purposes only and should not be used in production environments as they may be discontinued at short notice". Llama 3.3 70B and Llama 3.1 8B left the free and developer plans on 2026-08-16 and are Enterprise-only. Orpheus text-to-speech (English and Saudi Arabic) is also on the free plan, at 10 requests a minute, 100 requests and 3,600 tokens a day. Developer plan prices per 1M tokens: gpt-oss-120b $0.15/$0.60; gpt-oss-20b $0.075/$0.30; Qwen3.8 27B $0.80/$4.00. The Batch API costs 50% less and returns results within 24 hours to 7 days, without impacting standard rate limits. When requests to gpt-oss-20b, gpt-oss-120b or gpt-oss-safeguard-20b start with the same text as a recent request, Groq caches that part automatically and bills those input tokens 50% less. The batch discount does not stack with prompt caching discounts; all batch tokens are billed at the batch rate regardless of cache status. Orpheus text-to-speech costs $22 per 1M characters in English and $40 in Saudi Arabic. Groq's API is OpenAI-compatible at the base URL https://api.groq.com/openai/v1. The free models' API model IDs are openai/gpt-oss-120b, openai/gpt-oss-20b and qwen/qwen3.8-27b. Note — 2026-10-06: Groq cut the free limits of gpt-oss-safeguard-20b from 30 to 3 requests a minute and from 8,000 to 2,000 tokens a minute; 1,000 requests and 200,000 tokens a day are unchanged. The Internet Archive's capture of console.groq.com/docs/rate-limits on 2026-09-28 lists 30 RPM and 8K TPM, and its capture of 2026-10-06 lists 3 RPM and 2K TPM. Groq announced no change. The other free chat models keep 30 requests and 8,000 tokens a minute. Source ↗

• Groq rate limits apply to the whole organization, not per user or API key. A request over the limit returns a 429 Too Many Requests error. Cached tokens do not count toward rate limits. (From console.groq.com/docs/rate-limits, read 2026-11-04.)
• You are billed only after you add a payment method and move to the paid Developer tier. (From console.groq.com/docs/billing-faqs, read 2026-10-05.)
• Groq's Services Agreement does not permit it to train or fine-tune models on your inputs or outputs unless you give permission. (From console.groq.com/docs/legal/services-agreement, read 2026-10-05.)
• Groq's Services Agreement grants use of Groq APIs to integrate its services into customer applications for end users. Its restrictions do not prohibit commercial use. (From console.groq.com/docs/legal/services-agreement, read 2026-10-29.)
• Customers may not resell access to their Groq account. Groq's Services Agreement prohibits selling, reselling, or transferring its services except with Groq's express approval. (From console.groq.com/docs/legal/services-agreement, read 2026-10-29.)
• Groq does not retain inference requests by default. Batch input and output files are kept for 30 days unless deleted earlier. Fine-tuning data is kept until deleted. Groq may log inference inputs and outputs for up to 30 days to troubleshoot errors or investigate abuse. Zero Data Retention stops this logging and disables the Batch API and fine-tuning. (From console.groq.com/docs/your-data, read 2026-11-04.)
• Data Groq retains is stored in Google Cloud buckets in the United States. (From console.groq.com/docs/your-data, read 2026-10-05.)

Labelled llm_api. We read that on 2026-09-01 from https://groq.com/pricing, where it says: “One fully integrated platform for infrastructure, inference, and control”. How we use this

Context and memory layer for AI agents: Zep gives an agent context from past conversations, user activity and changing preferences. The Free plan includes 10,000 credits a month, with no rollover or auto top-up. An Episode (a chat message, JSON payload or block of text) up to 350 bytes costs 1 credit; retrieval, storage and users cost none. Free allows 2 projects, 1 Memory MCP Server seat and 5 custom entity and edge types, with variable rate limits and lower-priority processing. When the credits run out, Episodes stop processing until the next billing cycle and the account's rate limit drops to 5 requests a minute. Flex, the first paid plan, is $125/month for 50,000 credits. The self-hosted Community Edition is deprecated; Graphiti, the framework under Zep, is open source (Apache 2.0).

• Zep's terms (last updated 2026-08-17) give Zep a perpetual, irrevocable licence to use customer data on every plan, Free included, to train and improve AI models; the terms name no opt-out. (From getzep.com/legal/terms, read 2026-10-12.)
• The terms' data-security commitments, including safeguards for customer data and notice of a security incident, cover Paid plans only, not Free. (From getzep.com/legal/terms, read 2026-10-12.)

Labelled agent_memory. We read that on 2026-10-09 from https://www.getzep.com/, where it says: “Give agents context from past conversations, user activity, and changing preferences.”. How we use this

Memory and search API for AI agents, with user profiles and plugins for coding agents. The Free plan is $0/month with $5 of credits included, renewed every month; plan credits do not roll over. Credits pay the same rates on every plan, such as $5 per 1M SM tokens of plain-text memory and $5 per 1M search queries, and usage needs an available credit balance. Free includes the memory and search API and the coding-agent plugins for 1 team seat and unlimited end users; connectors such as Google Drive and Notion need a paid plan. Pro is $19/month with $20 of credits included.

Labelled agent_memory. We read that on 2026-10-09 from https://supermemory.ai/, where it says: “supermemory is building the default engine for memory and continual learning for agents.”. How we use this

Memory layer for AI agents and apps. Free plan: unlimited end users and community support. For the first 30 days Mem0 pays for the model calls, with 10,000 adds and 1,000 retrievals a month. After that, adding memories needs your own LLM key (OpenAI, Anthropic, Azure OpenAI or OpenRouter), billed by your provider; Mem0 puts its recommended models at about $3.55 to $5.34 per 1,000 adds. With a key, Free allows 50,000 adds and 10,000 retrievals a month at no charge from Mem0. Without one, stored memories stay and up to 1,000 searches a month still work, but new adds return an error. Calls past a monthly limit return an error until the next month, unless you connect a key and turn on usage-based billing with a monthly cap you set (Mem0 shows the rate when you turn it on). Mem0 Free accounts that signed up before October 26, 2026 have model calls paid by Mem0 until November 25, 2026; after that, adding memories requires their own LLM key. Starter ($19/month), Growth and Pro are no longer sold to new accounts; existing subscribers keep their plans. Graph memory, user profiles, SSO and on-prem deployment are on the Custom plan, priced by Mem0's sales team. Under Mem0's terms (2026-08-03), Free Plan users grant Mem0 a perpetual licence to their content, including to train AI models; paid plans exclude that, apart from aggregated, de-identified data. Mem0 Open-Source (Apache 2.0) can be self-hosted. Note — 2026-10-26: On October 26, 2026, Mem0 stopped selling its Starter ($19/month), Growth and Pro plans to new accounts and moved to two plans, Free and Custom. On Free, Mem0 pays for model calls only for the first 30 days (10,000 adds and 1,000 retrievals a month); after that, adding memories needs your own LLM key, with which Free allows 50,000 adds and 10,000 retrievals a month. Mem0 Free accounts that signed up before October 26, 2026 have their model calls paid by Mem0 until November 25, 2026. Source ↗

Labelled agent_memory. We read that on 2026-10-09 from https://mem0.ai/pricing, where it says: “Drop-in memory infrastructure for AI agents and apps.”. How we use this

Integration platform for AI Agents and LLMs — Free $0 No credit card required 100,000 tool calls / mo 50,000 triggers / mo 3 team members

• On the free Hobby plan, calls stop at the monthly limit and Composio does not charge overage. (From composio.dev/pricing/platform, read 2026-11-11.)
• Of the 100,000 free tool calls a month, up to 20,000 can use Composio-managed apps; calls through your own app, API key or MCP server can use the full allowance. (From composio.dev/pricing/platform, read 2026-11-11.)
• Of the 50,000 free trigger events a month, up to 10,000 can use Composio-managed apps. (From composio.dev/pricing/platform, read 2026-11-11.)

Labelled agent_tool_access. We read that on 2026-09-01 from https://composio.dev/, where it says: “secure delegated auth, sandboxed environments, and parallel execution across 1,000+ apps”. How we use this

$5 in free credits every month, applied automatically. Brave's "up to $20" figure is $5 on each of four plans: Search, Answers, Spellcheck and Autosuggest. Search is $5 per 1,000 requests; Answers is $4 per 1,000 requests plus $5 per million tokens. Web search, news, videos, images, LLM context endpoints. 50 queries/sec. Credit card required. The earlier free plan (2,000 queries/mo) ended Feb 2026. The tools of Brave's MCP server include brave_web_search, brave_news_search, brave_image_search, brave_llm_context and brave_summarizer.

• Keeping any data from the API is prohibited. Storing results or training an LLM requires an arrangement with Brave. (From api-dashboard.search.brave.com/documentation/resources/help-feedback, read 2026-10-06.)
• Once the free $5 and any bought credits are used up, requests are rejected until funds are added. Auto-reload is optional. (From api-dashboard.search.brave.com/documentation/resources/help-feedback, read 2026-10-06.)
• Unused free monthly credit does not carry over. (From api-dashboard.search.brave.com/documentation/resources/help-feedback, read 2026-10-06.)
• Brave's Service Credit Terms say credits, bought or free, expire one year after they are issued or when the account closes, while its help page says bought credits do not expire. (From api-dashboard.search.brave.com/terms-of-service/credit-terms, read 2026-10-07.)
• Bought credits are not refunded. (From api-dashboard.search.brave.com/terms-of-service/credit-terms, read 2026-10-07.)
• Brave's February launch post says the $5 monthly credit requires attributing the Brave Search API on the project's website, while Brave's Terms of Use (updated September 1, 2026) make attribution optional. Brave's help page describes getting the credit with a card at signup and does not mention attribution. (From api-dashboard.search.brave.com/terms-of-service, read 2026-10-11.)
• Brave keeps a record of search queries sent through an API account for up to 90 days for billing and troubleshooting. Zero data retention is available only for Enterprise customers. (From api-dashboard.search.brave.com/documentation/resources/privacy-notice, read 2026-10-11.)
• Brave's Search API Terms of Use license the API for the customer's own applications, products or services. Free credits are under the same Terms. The customer may not make the API available to third parties or get around rate limits or service limits, including by creating multiple accounts. (From api-dashboard.search.brave.com/terms-of-service, read 2026-10-30.)

Labelled web_search_api. We read that on 2026-10-19 from https://brave.com/search/api/, where it says: “The real-time search data your chatbots & agents need to generate answers.”. How we use this

Web Search API for AI agents. 100 free queries a day with no API key or account through the keyless MCP profile (api.you.com/mcp?profile=free), limited to the you-search and you-discover tools. New API accounts also get a $100 sign-up credit with no credit card required; no page says it renews or expires. Then pay as you go with no minimums: Web Search API $5.00 per 1,000 calls with up to 100 results per call, live page extraction $1.00 per 1,000 pages, at 10 requests a second on self-serve accounts.

Labelled web_search_api. We read that on 2026-09-27 from https://you.com/llms.txt, where it says: “Returns real-time web and news results as LLM-ready JSON.”. How we use this

Real-time web search API for AI agents, with search, extract, crawl and research endpoints. The free Researcher plan gives 1,000 API credits per month with no credit card required, resetting on the first of each month. A basic search costs 1 credit and an advanced search 2; extract costs 1 credit per 5 successful URLs (2 at advanced depth); research costs 4 to 250 credits per request. Search and Extract can also be called without an API key, free and rate-limited. Pay-as-you-go costs $0.008 per credit; the Project plan is $30 per month for 4,000 credits.

• When monthly free credits are used up, requests stop until credits reset or the account upgrades. Pay-as-you-go charging applies if enabled. (From tavily.com/pricing, read 2026-10-06.)
• Production API keys need an active paid plan or pay-as-you-go enabled. Development keys are limited to 100 requests per minute, production keys 1,000. (From docs.tavily.com/documentation/rate-limits, read 2026-10-10.)
• Tavily may use your query data to improve its services; an account setting turns this off, and then query data is not stored or used for improvements. (From help.tavily.com/articles/4205958832-understanding-the-allow-use-of-query-data-setting, read 2026-10-06.)
• Tavily for Startups gives up to 100,000 API credits for 6 months to accepted startups: early-stage, VC-backed companies building AI agents or LLMs, typically pre-seed through Series A. (From tavily.com/startups, read 2026-10-16.)
• Students get 4 months of the Project plan (4,000 API credits a month) free after verifying with a student email address or student ID. (From help.tavily.com/articles/6606514713-student-account, read 2026-10-16.)

Labelled web_search_api. We read that on 2026-09-01 from https://tavily.com/, where it says: “Fast and secure APIs for web search and content extraction”. How we use this

Search and research API for AI agents: one credit balance covers Search, Contents and DeepResearch. Accounts without a plan get free credits every month, which reset and do not roll over: 2,000 with a work email, no card needed; 1,000 with a personal or university email, which needs a card on the account from the second month (adding one charges nothing; one card unlocks free credits for one account, and prepaid cards don't count). Free monthly credits cover web search, arXiv and PubMed; other sources need a plan. A web search returning ten results uses 1 credit. Then pay as you go: $10 buys 1,600 credits, which last 12 months; Ridiculously Cheap is $29 a month for 5,000 credits. There are no overage charges: a request the balance cannot cover gets HTTP 402.

• In the first month only, the first 1,000 free credits with a work email (500 with a personal email) also work on every source on Valyu's standard plans, such as filings, trials or patents. Valyu's own pricing page at www.valyu.ai states the 1,000 without the first-month limit or the personal-email figure. (From docs.valyu.ai/pricing, read 2026-11-12.)
• Valyu's free credits were a one-time grant until late 2026: its changelog entry for October 2026 gives new accounts $10 of credits, or $20 with a company email, usable on every standard-plan source. Its pricing page now gives a monthly allowance that resets each month. (From docs.valyu.ai/pricing, read 2026-11-12.)

Labelled web_search_api. We read that on 2026-11-12 from https://docs.valyu.ai/llms.txt, where it says: “One API finds evidence in filings, papers, trials, patents, market data and the web, and returns it with citations.”. How we use this

Web search API for AI agents with Search, Contents, Answer, Monitors and Agent endpoints. The free tier gives $10 in credits each month, up to 2,500 Lite searches, resetting on the first of the month without rollover, plus a one-time $10 onboarding bonus; no credit card or other payment method required. Only a user's first team gets the free credits. Free teams get 10 QPS on search. Search costs $4 per 1,000 requests with up to 10 results for Lite, $7 for Auto, and $12 for Deep. Contents costs $1 per 1,000 pages per content type. Since October 19, 2026, Instant, Fast and Deep Reasoning are legacy search types that keep working at their current prices. Exa recommends Lite in place of Instant and Fast. Search and Contents can also be paid per request over x402 in USDC on Base or Solana, with no API key. Note — 2026-10-01: Between 2026-09-16 and 2026-10-01, Exa replaced the $20 sign-up credit with a one-time $10 bonus for completing onboarding, which only a user's first team earns. The $10 monthly credit, available in full on the day of sign-up, continues. Source ↗

• When a team's credits run out, requests return 402 Payment Required. Auto recharge is optional. (From exa.ai/docs/admin/billing, read 2026-10-06.)
• Exa uses anonymized data on default plans to improve its models. Zero Data Retention is available only on Enterprise plans. (From trust.exa.ai/faq, read 2026-10-06.)

Labelled web_search_api. We read that on 2026-09-01 from https://exa.ai/pricing, where it says: “Exa is a modern AI search engine with SERP API, website crawler tools, and deep research API”. How we use this

Real-time search engine scraping API. Returns structured JSON results for Google, YouTube, Bing, Baidu, Walmart, and many other engines. Free plan: 250 searches/month, 50 requests/hour throughput limit. Month-to-month, cancel anytime. See [serpapi.com/pricing](https://serpapi.com/pricing).

• SerpApi's U.S. Legal Shield, its cover for legal claims over scraping search engines, comes only with the Production plan and above, not the Free plan. (From serpapi.com/faq, read 2026-10-07.)

Labelled web_search_api. We read that on 2026-10-19 from https://serpapi.com/use-cases/web-search-api, where it says: “Turn raw web data into JSON or Markdown for your AI applications.”. How we use this

Web search API for AI agents. Accounts created with a professional email address get $20 of credit, topped back up to $20 each month (Linkup's docs); the pricing page calls this 4,000 free queries and states no period. Linkup's app asks other sign-ups to add a card, with no charge, to unlock the credit. Rate limit: 10 queries a second per organization. Then pay as you go from a prepaid balance: search $0.005 a call ($0.006 with a sourced answer or structured output), deep search $0.05 to $0.055, fetch from $0.001. Top-ups of $1,000 or more earn 10-20% bonus credit.

• Startups can apply for a one-time credit of up to $5,000 by telling Linkup what they are building. (From linkup.so/startups, read 2026-10-12.)
• When the account's credit runs out, the API responds with HTTP 429. (From docs.linkup.so/pages/documentation/platform/pricing, read 2026-10-12.)

Labelled web_search_api. We read that on 2026-09-27 from https://www.linkup.so/llms.txt, where it says: “Linkup is a web search API that gives AI applications and agents real-time, grounded access to the web through a single API call”. How we use this

Web search and research APIs for AI agents. Organizations with a credit card on file get $5 of free credit each month, and unused credit expires at month end. $5 covers up to 1,000 requests a month in Search's Basic or Advanced mode (Advanced is the API default), or 5,000 in Turbo or Fast mode; usage beyond $5 is billed to the card at standard rates. The $5 monthly credit applies to all Parallel APIs at their standard rates. Extract costs $1 per 1,000 URLs, and the Decisions API costs $1 per 1,000 requests, up to 100 questions per request. One monthly credit per organization and per card; marketplace and postpaid organizations are not eligible. The pricing page also offers "up to $80 at signup" without stating its terms. A hosted Search MCP is free without an account or API key, at lower rate limits, for personal and hobby use. Then pay as you go: Search $1 per 1,000 requests (Turbo, Fast) or $5 (Basic, Advanced) with 10 results, at 600 requests a minute.

Labelled web_search_api. We read that on 2026-09-27 from https://parallel.ai/products/search, where it says: “The highest accuracy web search API, built from the ground up for AIs”. How we use this

AI model API for Command chat models, Embed, Rerank, Transcribe and Parse. Every account starts with a Trial API key: calls made with it are free and limited to 1,000 API calls a month. Trial rate limits: 20 requests a minute per Chat model, 2,000 Embed inputs a minute and 10 Rerank requests a minute. Trial keys can use all of Cohere's models and APIs. Production keys are pay-as-you-go. Prices per 1M tokens: Command R7B $0.0375/$0.15; Command R $0.15/$0.60. The pricing page lists Command A+ (Apache 2.0) at $0 through an API key and as a model download.

• Cohere's Trial API keys may not be used for production or commercial purposes. (From cohere.com/pricing, read 2026-10-04.)
• Cohere creates a free Trial API key with each new account; no credit card is needed. (From cohere.com/pricing, read 2026-10-22.)

Labelled document_extraction. We read that on 2026-09-01 from https://cohere.com/pricing, where it says: “Parse New Document parsing model”. How we use this

20 free pages/month: Extract data from PDFs, emails. AI powered. Full API access.

Labelled document_extraction. We read that on 2026-09-01 from https://parseur.com, where it says: “automate text extraction from PDFs, emails, and other documents”. How we use this

Free OCR API that parses image and PDF files and returns the text as JSON: 25,000 requests/month, 1 MB file size limit, 3-page PDF limit, 500 requests/day per IP address, plus 1,000 Engine 3 conversions/month, down from 2,500 in October 2026. Engine 1 is retired; OCR.Space redirects calls to it to Engine 2. Free-tier searchable PDFs carry a watermark. PRO is $30/month for 300,000 requests and 5 MB files. The online OCR web form needs no registration and accepts files up to 5MB. Note — effective between 2026-10-03 and 2026-10-30: By October 30, 2026, OCR.Space's free OCR API plan includes 1,000 Engine 3 conversions a month, down from 2,500 on October 3. Engine 1 is retired, and OCR.Space redirects calls to it to Engine 2. The 25,000 monthly requests for Engines 1 and 2 are unchanged. Source ↗

Labelled document_extraction. We read that on 2026-09-01 from https://ocr.space/, where it says: “Web API to extract text from images and convert scans to searchable PDF”. How we use this

Serverless platform for running CPU and GPU workloads. The Starter plan has no monthly fee and includes $30 of compute per month (GPU, CPU and memory), 3 workspace seats, 100 containers and 10 GPU concurrency. Usage beyond the $30 is billed per second: T4 $0.000164 per second; H100 $0.001097 per second. Since 2026-09-01, Shared Endpoints are billed per token from the first request and are not covered by the $30. Since 2026-10-01, network egress beyond 1 TiB a month is billed at $0.04 per GiB. Team is $250 per month with $100 of compute included. Note — 2026-10-01: Since October 1, 2026, Modal charges for network egress: the Starter plan includes 1 TiB a month, Team 10 TiB and Enterprise 100 TiB, and egress beyond that costs $0.04 per GiB. Source ↗

• Modal's billing docs say a payment method must be on file to use Modal. (From modal.com/docs/guide/billing, read 2026-10-23.)
• Modal's terms say Modal will not train any AI model on customer data, or send customer data into large language models, without the customer's prior written consent. (From modal.com/legal/terms, read 2026-10-23.)

Labelled agent_sandbox. We read that on 2026-10-31 from https://modal.com/pricing, where it says: “Isolated environments for coding agents, RL rollouts, and untrusted code”. How we use this

Comet makes Opik, for LLM and agent observability and evaluation, and an MLOps platform for ML experiment tracking. Opik is open source and free to self-host. Opik Free Cloud allows up to 10 team members, 25k spans per month and 60-day data retention, with agent tracing and analysis, test suites and assertions, and Agent Playground. MLOps Free covers 1 platform user, with experiment tracking, dataset management and versioning, Model Registry and 100GB of data storage. Academics can apply for a free Pro plan.

Labelled experiment_tracking. We read that on 2026-09-01 from https://www.comet.com/site/pricing/, where it says: “Track and compare machine learning training runs Dataset management and versioning Model Registry”. How we use this

ML experiment tracking, model registry, and LLM app tracing and evaluation (Weave). W&B's hosted service is now sold as CoreWeave Forge. Its Free plan is $0/mo and designed for personal development, with up to 5 model seats, 5 GB/mo of storage and 1 GB/mo of Weave data ingestion. Pro starts at $60/month with a 30-day free trial and is for teams with fewer than 50 employees. Self-managed W&B Server needs a license; W&B's docs offer a free trial license, and CoreWeave's pricing page lists no free self-hosted plan.

Labelled experiment_tracking. We read that on 2026-09-01 from https://wandb.ai/site/, where it says: “Experiments Track and visualize your experiments”. How we use this

Simulate, evaluate, and observe your AI agents — The 'OSS' tier is free forever for developers, small teams, and self-managed deployments. It includes features like a drop-in replacement, OTel compatibility, metrics & traces, built-in observability, budget management, and more. A 14-day free trial of 'Bifrost Enterprise' is available.

Labelled model_gateway. We read that on 2026-09-01 from https://getmaxim.ai/, where it says: “The fastest, most resilient, enterprise-grade LLM, MCP, and agent gateway”. How we use this

Rebranded to Respan; keywordsai.co redirects to respan.ai. Free plan: full platform, 100k logs, 1k scores, 5 datasets, 2 evaluators, 5 prompts. No credit card required. We could not read the page we cite for this offer when we last looked, on 2026-10-31, so we cannot confirm these terms today.

Labelled model_gateway. We read that on 2026-09-02 from https://www.respan.ai, where it says: “Point the SDK you already use at one endpoint to reach 1,000+ models”. How we use this

AI model router. Free plan: 50 requests a day, no BYOK. Free models are capped at 20 requests a minute; accounts that have bought at least $10 of credits get 1,000 free-model requests a day. OpenRouter's pricing page says "25+ free models" and "4 free providers"; its model list (openrouter.ai/api/v1/models) had 17 model IDs ending in ":free" on 2026-10-20, and that list changes from week to week. The Standard plan (pay-as-you-go) charges a 5.5% fee on credit purchases and includes BYOK up to $25,000 of list-price inference a month with no fees, 5% after. The Business plan charges an 8% fee on credit purchases. It adds EU and US in-region routing: requests sent to eu.openrouter.ai or us.openrouter.ai stay with providers in that region. Free and Standard do not include in-region routing. No contract or sales call is needed. On 2026-08-19 OpenRouter announced it is joining Stripe. OpenRouter says it will continue to operate as it is, with the same name, product and roadmap, and that existing integrations do not change.

• OpenRouter's free models need no credit card. (From openrouter.ai/blog/tutorials/free-llm-apis-compared, read 2026-10-22.)
• A request over a free-model limit gets a 429 error with the message "Rate limit exceeded". The daily limit counts by UTC day. (From openrouter.ai/docs/api-reference/limits, read 2026-10-29.)
• The paid variant of a model has no platform-level request cap. (From openrouter.ai/docs/api-reference/limits, read 2026-10-29.)
• With a negative credit balance, requests can fail, including for free models; adding credits to bring the balance above zero restores access. (From openrouter.ai/docs/api-reference/limits, read 2026-10-29.)
• OpenRouter does not store your prompts or responses unless you opt in to Private Input & Output Logging or OpenRouter Use of Inputs/Outputs, or content is flagged as potential CSAM. Both opt-ins are off by default. OpenRouter Use of Inputs/Outputs lets OpenRouter use prompt and completion data to improve the product, for a 1% discount on model usage. OpenRouter stores request metadata without prompt or response content. (From openrouter.ai/docs/guides/privacy/data-collection, read 2026-10-30.)
• Each provider has its own data handling policies. OpenRouter works with providers so that prompts are not trained on wherever possible, but there are exceptions. Account settings control routing to providers that may train on your data, separately for paid and free models. If you opt out of training, OpenRouter will not route to those providers. (From openrouter.ai/docs/guides/privacy/provider-logging, read 2026-10-30.)

Labelled model_gateway. We read that on 2026-09-01 from https://openrouter.ai/pricing, where it says: “Pay only for what you use with access to 500+ AI models”. How we use this

AI media gateway: one OpenAI-compatible API to image and video generation models from several providers, billed at each provider's price with no markup, platform fee or subscription. One model, FLUX.1 [schnell] FP8, is listed free per image ("Free to try", served through Fireworks AI). New accounts also get a $1 credit with no card required; Lumenfall's terms say promotional credits typically expire after 90 days. After that you top up prepaid credits. Under Lumenfall's terms (2026-03-05), content sent through free features, including free API usage, may be used to train AI models.

Labelled model_gateway. We read that on 2026-09-01 from https://lumenfall.ai/, where it says: “unified access to the top image and video models across leading providers”. How we use this

LLM inference gateway. With a free token from dash.llm7.io: 1 request a second, 60 a minute, 250 an hour and 100,000 tokens per 24 hours, counting input and output tokens; LLM7 says free quotas may be reduced without notice. Free access covers turbo-tier models not marked usage-only, such as GLM-5.3-Flash and codestral-latest. Other models, including all image and speech-to-text models, need Pro ($12/mo) or a paid balance. LLM7's limits page no longer lists limits for requests without a token. A text request without a token still worked on 2026-10-19. Note — discovered 2026-10-07: The free tier limits have been significantly reduced. The free tier now allows 60 requests per minute and 100,000 tokens per 24 hours, whereas previously it allowed 10 requests per minute and 500,000 tokens per 24 hours without a token. The free tier with a token now allows 60 requests per minute and 100,000 tokens per 24 hours, whereas previously it allowed 40 requests per minute and 1,000,000 tokens per 24 hours. Source ↗

• LLM7's terms, updated October 15, 2026, say users keep their rights to prompts and outputs and grant LLM7 a licence to process them only to operate, secure, troubleshoot and improve the service. (From github.com/chigwell/llm7.io/blob/main/TERMS.md, read 2026-10-23.)
• The API answers HTTP 429 when a gateway or upstream rate limit is hit. (From docs.llm7.io/api-reference/openapi.json, read 2026-10-23.)

Labelled model_gateway. We read that on 2026-09-01 from https://llm7.io, where it says: “Connect to leading AI models with one endpoint”. How we use this

ML model hub. Since 2026-10-07, free accounts get no included Inference Providers credits: free users buy credits or upgrade to PRO to call Inference Providers, which serves 200+ models. PRO ($9/month) gets $2.00 a month. Free accounts get 100GB of private storage and best-effort public storage; substantial storage needs PRO, Team or Enterprise. Note — 2026-10-07: Hugging Face stopped giving free accounts included monthly Inference Providers credits; free users now have to buy credits or upgrade to PRO to call Inference Providers. Source ↗

• A free Hugging Face account needs purchased credits, which take a payment method, to call Inference Providers. (From huggingface.co/docs/inference-providers/pricing, read 2026-10-22.)

Labelled model_gateway. We read that on 2026-09-01 from https://huggingface.co/docs/inference-providers/pricing, where it says: “route requests through Hugging Face”. How we use this

Keyless Classic API: one image model and GPT-OSS 20B text, free, no sign-up, rate-limited per IP. Every other model needs an API key and Pollen, and free Pollen comes only from quests. No credit card is required. On 2026-10-11, the nine setup quests paid 3 Pollen in total, worth about $3 because 1 Pollen is worth about $1. The documented API for text, image, audio and video requires a GitHub sign-in and an API key. On the keyed API, free use draws from Quest Pollen, with no promise of future grants. Paid-only models, such as Claude models, DeepSeek V4 Pro and Gemini 4 Argon, need purchased Pollen. Gemini 3.8 Flash (google/gemini-3.8-flash) is paid-only on Pollinations, so Quest Pollen cannot pay for it. Kimi K3 (moonshotai/kimi-k3) is not paid-only. Kimi K3.1 Pro (moonshotai/kimi-k3.1-pro) was paid-only until 2026-11-09, when Pollinations made it Quest-accessible; Quest Pollen can now pay for it. To run Claude Code on free Quest Pollen, run "polli harness claude-code on", start Claude Code with "claude --settings ~/.pollinations/claude-code.json", and use a model that is not paid-only. Quest Pollen pays for any model that Pollinations' live catalog (enter.pollinations.ai/models) does not mark paid-only. On 2026-10-18, more than 30 text models were not paid-only, among them GPT-6 Sol, GPT-6.1 Sol, GPT-6 Luna, GLM 5.3, Kimi K3, DeepSeek V4 Flash and DeepSeek V4.1 Flash. The command "polli harness claude-code on" connects Claude Code to Pollinations and uses openai/gpt-6-sol unless you specify another model. Pollinations' harness documentation says the command writes "a Claude Code settings file" and: "Claude Code reads the file only when launched with --settings; plain claude keeps your native login and ~/.claude settings." (From https://github.com/pollinations/pollinations/blob/main/CODING_HARNESSES.md, read 2026-10-20.) GPT-6 Sol costs 1.5 Pollen per million input tokens and 7.5 Pollen per million output tokens. For requests over 272,000 prompt tokens, GPT-6 Sol costs 3 Pollen per million input tokens and 11.25 Pollen per million output tokens, and these rates apply to the whole request. GPT-5 Nano costs 0.0375 and 0.30. GPT-OSS 20B costs 0.05 and 0.18. NVIDIA Nemotron 3.5 Lightning costs 0.045 and 0.165. Adding --model <id> to "polli harness claude-code on" picks another model. The documented API is OpenAI-compatible at the base URL https://gen.pollinations.ai/v1, with an API key from https://enter.pollinations.ai/keys. Pollinations' MCP documentation gives this Claude Code command: "claude mcp add --transport http pollinations https://gen.pollinations.ai/mcp/pollinations --header 'Authorization: Bearer YOUR_KEY'". Its tools include generateImage, generateText, generateVideo, generateAudio, listModels and getBalance. "Calls use the same Pollen wallet as the Pollinations API." The "polli harness claude-code on" command runs Claude Code itself on a Pollinations model; the MCP server gives an agent Pollinations tools. Model IDs of some of the models Quest Pollen can pay for: openai/gpt-6-sol, openai/gpt-6.1-sol, openai/gpt-6-luna, z-ai/glm-5.3, moonshotai/kimi-k3, deepseek/deepseek-v4-flash, deepseek/deepseek-v4.1-flash, openai/gpt-5-nano, openai/gpt-oss-20b, nvidia/nemotron-3.5-lightning, moonshotai/kimi-k3.1-pro. Note — 2026-07-10: Pollinations ended the free Pollen that refilled every hour on every account (0.01 to 0.4 Pollen an hour, by tier). Free Pollen now comes only from Quests, whose rewards are paid once and land as Quest Pollen; further use needs purchased Pollen. The change did not touch the keyless image and text endpoints, which Pollinations documents as the free Classic API. Source ↗

• Pollinations' Classic API needs no sign-up and no key. image.pollinations.ai (one image model; JPEGs up to 768 × 768 with a small Pollinations logo) and text.pollinations.ai (GPT-OSS 20B, with an OpenAI-compatible endpoint at text.pollinations.ai/openai) are free and rate-limited per IP address: one image request at a time, at least 45 seconds apart, and one text request at a time, at least 30 seconds apart. Past the limit they return HTTP 402 with a Retry-After header. Pollinations runs it on a best-effort basis with no uptime guarantee, and says limits and models can change. Every other model needs a key from enter.pollinations.ai. (From github.com/pollinations/pollinations/blob/master/APIDOCS.md, read 2026-11-06.)
• Pollinations' API docs publish no fixed request rate for secret keys, which carry their own model permissions and budgets. A request can get HTTP 429 with a Retry-After header saying how long to wait. (From github.com/pollinations/pollinations/blob/main/APIDOCS.md, read 2026-10-23.)
• Pollinations' Terms of Service state that model licences vary and commercial use depends on each model's licence. The Terms set no separate rules for usage paid with free Quest Pollen. There is no SLA. (From pollinations.ai/terms, read 2026-10-30.)
• Pollinations does not use prompts or outputs to train or fine-tune models without explicit opt-in. Prompts and outputs are processed transiently and not retained beyond what is needed to run the service, except saved configuration and short-lived response caches. A request to a community-hosted model is sent to that provider, whose own terms govern storage and training; Pollinations does not verify those practices. (From pollinations.ai/privacy, read 2026-10-30.)
• Pollinations stores generated images, video, audio, and 3D files at media.pollinations.ai. Retrieval is public with the URL. These files have a 30-day lifecycle from upload or refresh. (From github.com/pollinations/pollinations/blob/main/APIDOCS.md, read 2026-11-03.)
• Pollinations requires browser apps to use a user-authorized sk_ key from Pollinations Pay. Personal sk_ keys must not be in browser code. A pk_ key only identifies the app and returns 401 if sent as a key. (From github.com/pollinations/pollinations/blob/main/APIDOCS.md, read 2026-11-03.)
• Pollinations Pay, also called BYOP, lets users authorize an app to spend Pollen from their own wallet. Users set budget, expiry, models, and permissions. They can edit or revoke issued keys. (From github.com/pollinations/pollinations/blob/main/enter.pollinations.ai/POLLEN_FAQ.md, read 2026-11-03.)
• Pollinations' image endpoint has a safe parameter. It defaults to off. The value nsfw enables sexual and violence filtering. The value true enables privacy and secrets filtering. (From github.com/pollinations/pollinations/blob/main/APIDOCS.md, read 2026-11-03.)
• Owners of public community models in Pollinations can set prices. Pollinations credits owners 75% of the Pollen spent on their models. A linked GitHub username is required. (From github.com/pollinations/pollinations/blob/main/enter.pollinations.ai/POLLEN_FAQ.md, read 2026-11-03.)

Labelled model_gateway. We read that on 2026-09-01 from https://github.com/pollinations/pollinations, where it says: “All models, one place: Flux, GPT, Claude, Gemini, Seedream, and more”. How we use this

ML platform (Gradient Notebooks, Workflows and Deployments) from Paperspace, now part of DigitalOcean. The Free plan is $0 with public projects, 5 GB of storage and free machines for Notebooks in your private workspace (Free CPU C4 and Free GPU M4000; free machines shut down after at most 6 hours). Other machines are billed per hour. Pro is $8/month for individuals and $12/month for teams; Growth is $39/month. DigitalOcean says new customers can still sign up but recommends its own GPU products instead.

Labelled gpu_compute. We read that on 2026-09-01 from https://www.paperspace.com/pricing, where it says: “Instances outside of the free-tier are billed per hour at their hourly rate”. How we use this

Hosted Jupyter notebooks with free access to GPUs and TPUs. In the free version, notebooks run for at most 12 hours, depending on availability and usage, and idle runtimes time out. GPU access is heavily restricted, GPU and TPU types vary over time, and usage limits fluctuate. Colab Pro, Pro+ and Pay As You Go offer more compute, based on a balance of compute units. The page we cite for Google Colab states in words that it is free, which confirms the price — the limits stated here come from our own record rather than from that page.

• On a free runtime with no compute-unit balance, Google may terminate without warning sessions that use SSH shells, remote desktops, a web UI instead of the notebook, chess training or distributed computing workers. SSH shells, remote desktops and working in a web UI instead of the notebook are allowed on a paid plan while the compute-unit balance is positive. (From research.google.com/colaboratory/faq.html, read 2026-10-08.)
• A paid Google AI plan adds a monthly allotment of Colab compute units and stronger GPUs and TPUs to the same account. Storage-only Google One plans and free trials do not. (From research.google.com/colaboratory/faq.html, read 2026-10-08.)
• Google AI Plus adds 50 Colab compute units a month, but not during a free trial of the plan. (From support.google.com/googleone/answer/16882689, read 2026-10-16.)
• Google AI Pro adds 200 Colab compute units a month, but not during a free trial of the plan. (From support.google.com/googleone/answer/14534406, read 2026-10-16.)
• Google AI Ultra adds 1,000 or 2,000 Colab compute units a month, depending on the plan, with longer background execution and premium GPU access, but not during a free trial of the plan. (From support.google.com/googleone/answer/16286513, read 2026-10-16.)
• Colab Pro for Education, a free year of Colab Pro for students and faculty at US universities, is closed to new signups. (From research.google.com/colaboratory/faq.html, read 2026-10-16.)
• A running notebook can be terminated if compute units run out. The account then uses the free tier's rules until more units are added. (From research.google.com/colaboratory/faq.html, read 2026-10-10.)
• Colab bans file hosting, media serving and other web services not related to interactive compute on every managed runtime, free or paid. (From research.google.com/colaboratory/faq.html, read 2026-10-10.)
• Google publishes no RAM figure for Colab. The free version gets VMs with a standard system memory profile. Paid plans can get a high memory system profile, subject to availability and the compute unit balance. (From research.google.com/colaboratory/faq.html, read 2026-10-23.)
• On Colab Pro+ with a positive compute unit balance, a running notebook keeps executing after the browser closes, for up to 24 hours. Idle VMs are still deleted. Colab Pro gets extended execution times based on availability. Without a paid subscription, execution stops when you stop interacting. (From colab.research.google.com/signup, read 2026-10-23.)
• Google Colab offers Pay As You Go at $9.99 for 100 Compute Units or $49.99 for 500 Compute Units, with no subscription. Colab Pro is $9.99 per month for 100 compute units. Colab Pro+ is $49.99 per month for 600 compute units. Compute units are not refundable, not transferable, and expire after 90 days. These are US prices. (From colab.research.google.com/signup, read 2026-11-09.)

Labelled gpu_compute. We read that on 2026-10-01 from https://research.google.com/colaboratory/faq.html, where it says: “provides free of charge access to computing resources, including GPUs and TPUs”. How we use this

Google's data science platform with free notebooks, datasets and competitions. Notebooks get a weekly GPU quota of 30 hours, sometimes more depending on demand (one NVIDIA Tesla P100 or two Tesla T4s), and a TPU v3-8 for up to 20 hours per week. Sessions run up to 12 hours on CPU or GPU and 9 hours on TPU, with 20 GB of auto-saved disk in /kaggle/working. A CPU session has 4 CPU cores and 30 GB of RAM. A GPU session, with one P100 or two T4s, has 4 CPU cores and 29 GB of RAM. A TPU v3-8 session has 96 CPU cores and 330 GB of RAM. Free GPUs and TPUs may queue at busy times. Datasets are capped at 200 GB each and 200 GB of private datasets in total. Kaggle's one paid service is hosting a competition. The tools of Kaggle's MCP server include save_notebook (run with GPU or TPU), get_notebook_session_status, download_notebook_output and get_accelerator_quota (read weekly GPU and TPU usage). The Kaggle API also pushes a notebook version without the web editor. (From https://www.kaggle.com/docs/efficient-gpu-usage, read 2026-10-23.)

• Kaggle's terms limit use of its services to your own internal, personal, non-commercial use, not on behalf of or for the benefit of any third party. (From kaggle.com/terms, read 2026-10-04.)
• Kaggle requires phone verification for parts of its platform, including notebooks, datasets, discussion topics and competitions. (From kaggle.com/discussions/product-announcements/558078, read 2026-10-04.)
• Linking an active Colab Pro or Pro+ subscription to a Kaggle account adds 15 or 30 GPU hours a week, respectively, as a limited-time promotion that Kaggle can end at any time. (From kaggle.com/docs/notebooks, read 2026-10-08.)
• Once Kaggle's weekly GPU quota is used, no more GPU time is available until the quota resets. The quota resets weekly. (From kaggle.com/docs/efficient-gpu-usage, read 2026-10-29.)
• Kaggle suggests Google Cloud AI Notebooks for users needing more compute. These notebooks run under your Google Cloud project and can be configured with your choice of virtual machine and accelerators, without Kaggle's limits. (From kaggle.com/docs/notebooks, read 2026-10-29.)
• Kaggle's interactive notebook sessions end after 20 minutes of idle time. Save & Run All runs the entire notebook in a separate session. That session must complete within 12 hours, or 9 hours for TPU notebooks. (From kaggle.com/docs/notebooks, read 2026-11-03.)
• New Kaggle notebooks have the Internet setting on by default. Pip installs are possible. Kaggle saves up to 20 GB of output in /kaggle/working. This data is reusable in later notebooks. Other disk space is scratch and not saved after the session. (From kaggle.com/docs/notebooks, read 2026-11-03.)
• New Kaggle notebooks are private by default. Datasets are Private by default. They are visible only to the owner, collaborators, and Kaggle for purposes consistent with its Privacy Policy. Public datasets are visible to everyone. (From kaggle.com/docs/datasets, read 2026-11-03.)

Labelled gpu_compute. We read that on 2026-09-01 from https://www.kaggle.com/docs/notebooks, where it says: “a cloud computational environment that enables reproducible and collaborative analysis”. How we use this

Serverless platform for running CPU and GPU workloads. The Starter plan has no monthly fee and includes $30 of compute per month (GPU, CPU and memory), 3 workspace seats, 100 containers and 10 GPU concurrency. Usage beyond the $30 is billed per second: T4 $0.000164 per second; H100 $0.001097 per second. Since 2026-09-01, Shared Endpoints are billed per token from the first request and are not covered by the $30. Since 2026-10-01, network egress beyond 1 TiB a month is billed at $0.04 per GiB. Team is $250 per month with $100 of compute included. Note — 2026-10-01: Since October 1, 2026, Modal charges for network egress: the Starter plan includes 1 TiB a month, Team 10 TiB and Enterprise 100 TiB, and egress beyond that costs $0.04 per GiB. Source ↗

• Modal's billing docs say a payment method must be on file to use Modal. (From modal.com/docs/guide/billing, read 2026-10-23.)
• Modal's terms say Modal will not train any AI model on customer data, or send customer data into large language models, without the customer's prior written consent. (From modal.com/legal/terms, read 2026-10-23.)

Labelled gpu_compute. We read that on 2026-10-31 from https://modal.com/pricing, where it says: “Fine-tune, run RL, and evaluate models on GPUs that spin up in seconds”. How we use this

AI model API for Command chat models, Embed, Rerank, Transcribe and Parse. Every account starts with a Trial API key: calls made with it are free and limited to 1,000 API calls a month. Trial rate limits: 20 requests a minute per Chat model, 2,000 Embed inputs a minute and 10 Rerank requests a minute. Trial keys can use all of Cohere's models and APIs. Production keys are pay-as-you-go. Prices per 1M tokens: Command R7B $0.0375/$0.15; Command R $0.15/$0.60. The pricing page lists Command A+ (Apache 2.0) at $0 through an API key and as a model download.

• Cohere's Trial API keys may not be used for production or commercial purposes. (From cohere.com/pricing, read 2026-10-04.)
• Cohere creates a free Trial API key with each new account; no credit card is needed. (From cohere.com/pricing, read 2026-10-22.)

Labelled embeddings_api. We read that on 2026-09-01 from https://cohere.com/pricing, where it says: “Embed Search and discovery model”. How we use this

Vector database. The Starter plan is free: up to 2 GB storage, 2M write units and 1M read units a month, 1 GB of egress a month (reads that return data stop at the cap until the next billing period), 5 indexes, 1 project and 2 users, on AWS us-east-1 only. Assistant includes 1 GB storage and 500k input, 300k output and 500k context tokens a month. Three embedding models include 5M tokens a month each; bge-reranker-v2-m3 includes 500 requests a month. Builder is $20/month flat.

Labelled embeddings_api. We read that on 2026-09-01 from https://pinecone.io/pricing, where it says: “Inference - Embedding llama-text-embed-v2 5M tokens/mo included”. How we use this

Free tier: Gemini 4 Hydrogen, Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, Gemini 3 Flash Preview, Gemini Embedding 2 and Gemma 4 are free of charge. Also free of charge are Gemini 3.8 Live (real-time voice), Gemini 3.5 Live Translate (translation), Gemini 3.5 Transcribe (speech-to-text), Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS (text-to-speech). Gemini Decision (gemini-decision-1), released 2026-11-11, is a multimodal decision model available through the Decisions API. It answers yes/no, choice and rubric questions with probabilities and generates zero output tokens, so it has no output price; input is free of charge on the free tier with 10,000 requests per day, and $0.10 per million input tokens on the paid tier. Gemini 4 Hydrogen (gemini-4-hydrogen), released 2026-11-12, is free of charge on the Gemini API free tier for input, output and context caching. On the paid tier, Gemini 4 Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens (output includes thinking tokens), with context caching at $0.075 (implicit caching only); on the Batch API, Gemini 4 Hydrogen is $0.75 input and $3.75 output. The other Gemini 4 models, Gemini 4 Carbon (gemini-4-carbon) and Gemini 4 Argon (gemini-4-argon), are paid only, as are Gemini 3.1 Pro Preview and Gemini Omni 1.2 Flash (gemini-omni-1.2-flash). Gemini 4 Carbon, which Google calls its most capable reasoning model, has been generally available since 2026-11-10 and is not available on the Gemini API free tier. Gemini 4 Argon, generally available since 2026-10-13, is not available on the Gemini API free tier. According to ai.google.dev/gemini-api/docs/google-ai-plans, "Gemini 4 Argon is not available on the Free plan." Gemini 4 Argon is available through the Gemini API on the paid tier. "Direct use of the Gemini API (such as using API keys or external applications) is billed and managed separately," per ai.google.dev/gemini-api/docs/google-ai-plans. Google AI Pro and Ultra provide access to Gemini 4 Argon in Google AI Studio, per ai.google.dev/gemini-api/docs/google-ai-plans. Gemini 4 Argon is in Antigravity on Google AI Ultra and paid Google AI Pro. In Google AI Studio, Gemini 4 Carbon requires a Google AI Ultra plan and is not available on the Free or Google AI Pro plans (aistudio.google.com/docs/google-ai-plans). In Antigravity, Gemini 4 Carbon is available on Google AI Ultra and Enterprise only (antigravity.google/docs/models). On a Google AI Pro trial, Argon is only available "on trial with eligible premium devices" (antigravity.google/docs/models). Claude Opus 5.5 and Claude Sonnet 5.5 are not included on Google AI Pro trials; trial users and those with memberships through third-party partners, Pixel Pass or Google Fi manage subscriptions through Google One. Google AI Pro via partner offers like Jio's does not include Gemini 4 Argon, Claude Sonnet 5.5 or Claude Opus 5.5 in Antigravity. Per a Google staff member on discuss.ai.google.dev (2026-10-21), "Those frontier models are restricted to directly billed (non-trial/non-promotional) Google AI Pro, Google AI Ultra and Enterprise plans." No Gemini image model has a free tier in the API. Nano Banana 2.1 (gemini-nano-banana-2.1) became generally available on 2026-10-06. It costs $0.0336 per 1K image, $0.0504 per 2K image and $0.113 per 4K image. The Batch API halves those prices. Nano Banana Pro costs $0.134 per 1K or 2K image and $0.24 per 4K image. Except for Gemini Decision, Google publishes no free-tier rate limits; each project's limits are shown in Google AI Studio. Developers report 429 errors showing 20 requests per day for Gemini 3.x Flash models and 500 requests per day for 3.x Flash-Lite models as of September 2026. Rate limits apply per project, not per API key. Daily request quotas reset at midnight Pacific time, per Google's rate-limits page. Since 2026-09-18 Google serves the Gemini 2.5 models only to users who have used them before, and points new projects to 3.5 Flash-Lite or 3.8 Flash. If your app calls gemini-3.7-flash, Google deprecated it on 2026-10-08. Requests to gemini-3.7-flash are routed automatically to gemini-3.8-flash. Google says to update the model string to gemini-3.8-flash. Google deprecated gemini-2.5-pro and gemini-2.5-flash on 2026-11-02. Both models shut down on 2026-12-01. Use them until that date. From that date, requests to gemini-2.5-pro route to gemini-3.8-flash and are billed at its rates. Requests to gemini-2.5-flash route to gemini-3.5-flash-lite and are billed at its rates. Update your model string before 2026-12-01 and review Google’s migration guides for breaking API changes. gemini-2.5-flash-lite is not deprecated. If your app calls Gemini 3.5 Flash (gemini-3.5-flash), Google deprecated it on 2026-10-08. Requests to gemini-3.5-flash are routed automatically to gemini-3.6-flash. Google says to update the model string to gemini-3.6-flash or gemini-3.8-flash. Google shut down Gemini Omni Flash Preview (gemini-omni-flash-preview) on 2026-10-22. Google's replacement is gemini-omni-1.2-flash. Google shuts down Gemini Omni 1.1 Flash (gemini-omni-1.1-flash) on December 3, 2026; after that, requests that name gemini-omni-1.1-flash are answered by gemini-omni-1.2-flash at Gemini Omni 1.2 Flash rates. Gemini 3.1 Flash-Lite (gemini-3.1-flash-lite) can be called until 2027-05-07, when Google shuts it down. Google's replacement is gemini-3.5-flash-lite. Paid, per million tokens (input/output; output prices include thinking tokens): Gemini 4 Carbon $2/$10 through 2026-11-30 (context caching $0.10, then $0.20, implicit caching only); Gemini 4 Argon $2/$10 through 2026-11-30 (context caching $0.10, then $0.20, implicit caching only); Gemini 3.8 Flash $0.75/$3.75 until 2026-12-31, then $1.50 input and $7.50 output from 2027-01-01 (3.6 Flash: the same prices and dates); Gemini 3.5 Flash-Lite $0.30/$2.50; Gemini 3.1 Flash-Lite $0.25/$1.50; Gemini 3.1 Pro Preview $2/$12 (prompts up to 200K tokens; prompts over 200K tokens cost $4 input and $18 output). Through 2026-11-30, Gemini 4 Argon's Standard output price is below Gemini 3.1 Pro Preview's for prompts up to 200K tokens, with the same input price; from 2026-12-01 Argon's is higher. From December 1, 2026, Gemini 4 Argon will cost $4 per million input tokens and $20 per million output tokens for Standard, $2 input and $10 output for Batch API, $2 input and $10 output for Flex, and $7.20 input and $36 output for Priority. From December 1, 2026, Gemini 4 Carbon will also cost $4 per million input tokens and $20 per million output tokens for Standard, $2 input and $10 output for Batch API and Flex, and $7.20 input and $36 output for Priority. The Batch API costs half the standard price; Gemini 3.8 Flash is $0.375 input and $1.875 output per million tokens through 2026-12-31. Gemini 4 Carbon and Gemini 4 Argon on the Batch API are $1 input and $5 output through 2026-11-30. Google also sells two tiers, Flex and Priority, both in Preview. A request picks one with the service_tier field ("flex" or "priority"); without it the request is Standard. Gemini 4 Carbon and Gemini 4 Argon on Flex are $1 input and $5 output through 2026-11-30; on Priority they are $3.60 input and $18 output through 2026-11-30. "The Gemini Flex API is an inference tier that offers a 50% cost reduction compared to standard rates, in exchange for variable latency and best-effort availability" (ai.google.dev/gemini-api/docs/flex-inference). "Priority inference is priced at 75-100% more than the standard API and billed per token" (ai.google.dev/gemini-api/docs/priority-inference). The Gemini API Contributor Tier is an invite-only private preview. Eligible users will be invited by email. It offers half the Standard price for Gemini 4 Carbon, Gemini 4 Argon, Gemini 3.1 Pro Preview, Gemini 3.8 Flash, and Gemini 3.5 Flash-Lite. Google uses the data from those models to improve its products. "Please don't submit sensitive or personal information" (ai.google.dev/gemini-api/docs/logs-policy). Grounding with Google Search is supported on the free tier for Gemini 3.8 Live, Gemini 3.8 Live Extended Thinking and Gemini 3.1 Flash Live Preview. It is not available on the free tier for Gemini 4 Hydrogen, Gemini 4 Argon or the Gemini 3.x Flash, Flash-Lite and Pro models. On the paid tier, 5,000 search requests a month are free, shared across Gemini 3 and newer models, then $14 per 1,000. Accounts opened after 2026-03-02 cannot spend the $300 Google Cloud welcome credit on the Gemini API.

• The Gemini API Free Tier needs no credit card or billing account; linking a billing account and prepaying moves you to the Paid Tier. (From ai.google.dev/gemini-api/docs/billing, read 2026-10-22.)
• Google's pricing page says free-tier content is used to improve Google's products. (From ai.google.dev/gemini-api/docs/pricing, read 2026-10-04.)
• For developers outside the European Economic Area, Switzerland and the United Kingdom, Google's terms say human reviewers may read free-tier prompts and responses, and Google says not to send sensitive, confidential or personal information on the free tier. (From ai.google.dev/gemini-api/terms, read 2026-10-04.)
• Paid-tier projects created on or after October 26, 2026, store Generate Content API requests and responses by default. Projects created before that date keep their existing Generate Content API storage setting, which is off unless it was turned on in AI Studio. The Interactions API stores requests by default in all projects, including projects created before October 26, 2026. The Paid Tier keeps stored interactions for 55 days and the Free Tier for 1 day. Logging also covers calls made through OpenAI compatibility endpoints. Logs are retained for up to 55 days, configurable to 7, 14 or 28 days. Logs are not used to improve Google's products unless shared. Logging can be toggled independently for each API in AI Studio, and store=false on a request turns it off for that request. (From ai.google.dev/gemini-api/docs/logs-policy, read 2026-11-06.)
• Apps that serve users in the European Economic Area, Switzerland or the United Kingdom may use only the paid tier. (From ai.google.dev/gemini-api/terms, read 2026-10-04.)
• Google's terms require users to be 18 years of age or older to use the Gemini API and Google AI Studio. The API may not be used in a website, app, or other service directed towards or likely to be accessed by individuals under 18. These rules apply to both the free and paid tiers. (From ai.google.dev/gemini-api/terms, read 2026-10-17.)
• Google offers the Gemini API and Google AI Studio, free and paid, in 230 countries and territories, including Taiwan, Brazil, Japan, Singapore, and the United Kingdom. China, Hong Kong, Macao, Russia, Belarus, and Iran are not among them. Google says to developers outside the list: "try the Gemini API in Gemini Enterprise Agent Platform". (From ai.google.dev/gemini-api/docs/available-regions, read 2026-10-17.)
• Google's terms say changes to Gemini API paid prices take effect 30 days after Google posts them unless Google specifies otherwise. Pricing for a new paid service takes effect immediately unless Google specifies otherwise. (From ai.google.dev/gemini-api/terms, read 2026-10-17.)
• Google is moving existing Postpay accounts to Prepay for Gemini API usage. An account that has not switched and added credits by the cutover date in its account notice has its paid service interrupted; other Google Cloud services on the billing account stay on Postpay. (From ai.google.dev/gemini-api/docs/billing, read 2026-10-07.)
• When the Prepay balance reaches $0, every API key in every project on the billing account stops working until credits are added; requests then fail with an HTTP 402 Payment Required error. (From ai.google.dev/gemini-api/docs/billing, read 2026-10-07.)
• The monthly Google Cloud credits from Google AI Pro or Ultra are eligible Google Cloud credits. On the Prepay billing plan, these credits pay for Gemini API usage only after you buy Prepay credits in AI Studio. With a $0 Prepay balance, requests fail even with unused Google Cloud credits, according to a Google staff member on discuss.ai.google.dev. (From discuss.ai.google.dev/t/error-code-402-message-your-prepayment-credits-are-depleted/185524, read 2026-10-10.)
• A paid project does not automatically revert to the free tier, even when its Prepay balance reaches $0. To use the free tier, disable billing on the project, which unlinks it from its billing account. (From ai.google.dev/gemini-api/docs/billing, read 2026-10-10.)
• Disabling billing on a Google Cloud project stops all other paid Google Cloud services in it. Some resources may be deleted and become non-recoverable. Outstanding charges are still due. (From cloud.google.com/billing/docs/how-to/modify-project, read 2026-10-10.)

Labelled embeddings_api. We read that on 2026-09-01 from https://ai.google.dev/pricing, where it says: “Our first multimodal embedding model, mapping text, images, video, audio, and PDFs into a unified embedding space”. How we use this

Free of charge text, image, audio, and video input on the Gemini API free tier, with Google using the content to improve its products. Batch is not available on the free tier. Paid rates are $0.20 per 1M text tokens, $0.45 per 1M image tokens, $6.50 per 1M audio tokens, and $12.00 per 1M video tokens. Batch paid rates are $0.10 per 1M text tokens, $0.225 per 1M image tokens, $3.25 per 1M audio tokens, and $6.00 per 1M video tokens. The same paid rates apply on Google's Gemini Enterprise Agent Platform (formerly Vertex AI). It embeds text, images, video, audio, and PDFs into one embedding space.

Labelled embeddings_api. We read that on 2026-09-01 from https://cloud.google.com/vertex-ai/docs/generative-ai/embeddings/get-text-embeddings, where it says: “Generate text embeddings with Agent Platform Text Embeddings API”. How we use this

AI model API for Command chat models, Embed, Rerank, Transcribe and Parse. Every account starts with a Trial API key: calls made with it are free and limited to 1,000 API calls a month. Trial rate limits: 20 requests a minute per Chat model, 2,000 Embed inputs a minute and 10 Rerank requests a minute. Trial keys can use all of Cohere's models and APIs. Production keys are pay-as-you-go. Prices per 1M tokens: Command R7B $0.0375/$0.15; Command R $0.15/$0.60. The pricing page lists Command A+ (Apache 2.0) at $0 through an API key and as a model download.

• Cohere's Trial API keys may not be used for production or commercial purposes. (From cohere.com/pricing, read 2026-10-04.)
• Cohere creates a free Trial API key with each new account; no credit card is needed. (From cohere.com/pricing, read 2026-10-22.)

Labelled speech_api. We read that on 2026-09-01 from https://cohere.com/pricing, where it says: “Transcribe New Speech recognition model”. How we use this

Free tier: Gemini 4 Hydrogen, Gemini 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, 3.1 Flash-Lite, Gemini 3 Flash Preview, Gemini Embedding 2 and Gemma 4 are free of charge. Also free of charge are Gemini 3.8 Live (real-time voice), Gemini 3.5 Live Translate (translation), Gemini 3.5 Transcribe (speech-to-text), Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS (text-to-speech). Gemini Decision (gemini-decision-1), released 2026-11-11, is a multimodal decision model available through the Decisions API. It answers yes/no, choice and rubric questions with probabilities and generates zero output tokens, so it has no output price; input is free of charge on the free tier with 10,000 requests per day, and $0.10 per million input tokens on the paid tier. Gemini 4 Hydrogen (gemini-4-hydrogen), released 2026-11-12, is free of charge on the Gemini API free tier for input, output and context caching. On the paid tier, Gemini 4 Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens (output includes thinking tokens), with context caching at $0.075 (implicit caching only); on the Batch API, Gemini 4 Hydrogen is $0.75 input and $3.75 output. The other Gemini 4 models, Gemini 4 Carbon (gemini-4-carbon) and Gemini 4 Argon (gemini-4-argon), are paid only, as are Gemini 3.1 Pro Preview and Gemini Omni 1.2 Flash (gemini-omni-1.2-flash). Gemini 4 Carbon, which Google calls its most capable reasoning model, has been generally available since 2026-11-10 and is not available on the Gemini API free tier. Gemini 4 Argon, generally available since 2026-10-13, is not available on the Gemini API free tier. According to ai.google.dev/gemini-api/docs/google-ai-plans, "Gemini 4 Argon is not available on the Free plan." Gemini 4 Argon is available through the Gemini API on the paid tier. "Direct use of the Gemini API (such as using API keys or external applications) is billed and managed separately," per ai.google.dev/gemini-api/docs/google-ai-plans. Google AI Pro and Ultra provide access to Gemini 4 Argon in Google AI Studio, per ai.google.dev/gemini-api/docs/google-ai-plans. Gemini 4 Argon is in Antigravity on Google AI Ultra and paid Google AI Pro. In Google AI Studio, Gemini 4 Carbon requires a Google AI Ultra plan and is not available on the Free or Google AI Pro plans (aistudio.google.com/docs/google-ai-plans). In Antigravity, Gemini 4 Carbon is available on Google AI Ultra and Enterprise only (antigravity.google/docs/models). On a Google AI Pro trial, Argon is only available "on trial with eligible premium devices" (antigravity.google/docs/models). Claude Opus 5.5 and Claude Sonnet 5.5 are not included on Google AI Pro trials; trial users and those with memberships through third-party partners, Pixel Pass or Google Fi manage subscriptions through Google One. Google AI Pro via partner offers like Jio's does not include Gemini 4 Argon, Claude Sonnet 5.5 or Claude Opus 5.5 in Antigravity. Per a Google staff member on discuss.ai.google.dev (2026-10-21), "Those frontier models are restricted to directly billed (non-trial/non-promotional) Google AI Pro, Google AI Ultra and Enterprise plans." No Gemini image model has a free tier in the API. Nano Banana 2.1 (gemini-nano-banana-2.1) became generally available on 2026-10-06. It costs $0.0336 per 1K image, $0.0504 per 2K image and $0.113 per 4K image. The Batch API halves those prices. Nano Banana Pro costs $0.134 per 1K or 2K image and $0.24 per 4K image. Except for Gemini Decision, Google publishes no free-tier rate limits; each project's limits are shown in Google AI Studio. Developers report 429 errors showing 20 requests per day for Gemini 3.x Flash models and 500 requests per day for 3.x Flash-Lite models as of September 2026. Rate limits apply per project, not per API key. Daily request quotas reset at midnight Pacific time, per Google's rate-limits page. Since 2026-09-18 Google serves the Gemini 2.5 models only to users who have used them before, and points new projects to 3.5 Flash-Lite or 3.8 Flash. If your app calls gemini-3.7-flash, Google deprecated it on 2026-10-08. Requests to gemini-3.7-flash are routed automatically to gemini-3.8-flash. Google says to update the model string to gemini-3.8-flash. Google deprecated gemini-2.5-pro and gemini-2.5-flash on 2026-11-02. Both models shut down on 2026-12-01. Use them until that date. From that date, requests to gemini-2.5-pro route to gemini-3.8-flash and are billed at its rates. Requests to gemini-2.5-flash route to gemini-3.5-flash-lite and are billed at its rates. Update your model string before 2026-12-01 and review Google’s migration guides for breaking API changes. gemini-2.5-flash-lite is not deprecated. If your app calls Gemini 3.5 Flash (gemini-3.5-flash), Google deprecated it on 2026-10-08. Requests to gemini-3.5-flash are routed automatically to gemini-3.6-flash. Google says to update the model string to gemini-3.6-flash or gemini-3.8-flash. Google shut down Gemini Omni Flash Preview (gemini-omni-flash-preview) on 2026-10-22. Google's replacement is gemini-omni-1.2-flash. Google shuts down Gemini Omni 1.1 Flash (gemini-omni-1.1-flash) on December 3, 2026; after that, requests that name gemini-omni-1.1-flash are answered by gemini-omni-1.2-flash at Gemini Omni 1.2 Flash rates. Gemini 3.1 Flash-Lite (gemini-3.1-flash-lite) can be called until 2027-05-07, when Google shuts it down. Google's replacement is gemini-3.5-flash-lite. Paid, per million tokens (input/output; output prices include thinking tokens): Gemini 4 Carbon $2/$10 through 2026-11-30 (context caching $0.10, then $0.20, implicit caching only); Gemini 4 Argon $2/$10 through 2026-11-30 (context caching $0.10, then $0.20, implicit caching only); Gemini 3.8 Flash $0.75/$3.75 until 2026-12-31, then $1.50 input and $7.50 output from 2027-01-01 (3.6 Flash: the same prices and dates); Gemini 3.5 Flash-Lite $0.30/$2.50; Gemini 3.1 Flash-Lite $0.25/$1.50; Gemini 3.1 Pro Preview $2/$12 (prompts up to 200K tokens; prompts over 200K tokens cost $4 input and $18 output). Through 2026-11-30, Gemini 4 Argon's Standard output price is below Gemini 3.1 Pro Preview's for prompts up to 200K tokens, with the same input price; from 2026-12-01 Argon's is higher. From December 1, 2026, Gemini 4 Argon will cost $4 per million input tokens and $20 per million output tokens for Standard, $2 input and $10 output for Batch API, $2 input and $10 output for Flex, and $7.20 input and $36 output for Priority. From December 1, 2026, Gemini 4 Carbon will also cost $4 per million input tokens and $20 per million output tokens for Standard, $2 input and $10 output for Batch API and Flex, and $7.20 input and $36 output for Priority. The Batch API costs half the standard price; Gemini 3.8 Flash is $0.375 input and $1.875 output per million tokens through 2026-12-31. Gemini 4 Carbon and Gemini 4 Argon on the Batch API are $1 input and $5 output through 2026-11-30. Google also sells two tiers, Flex and Priority, both in Preview. A request picks one with the service_tier field ("flex" or "priority"); without it the request is Standard. Gemini 4 Carbon and Gemini 4 Argon on Flex are $1 input and $5 output through 2026-11-30; on Priority they are $3.60 input and $18 output through 2026-11-30. "The Gemini Flex API is an inference tier that offers a 50% cost reduction compared to standard rates, in exchange for variable latency and best-effort availability" (ai.google.dev/gemini-api/docs/flex-inference). "Priority inference is priced at 75-100% more than the standard API and billed per token" (ai.google.dev/gemini-api/docs/priority-inference). The Gemini API Contributor Tier is an invite-only private preview. Eligible users will be invited by email. It offers half the Standard price for Gemini 4 Carbon, Gemini 4 Argon, Gemini 3.1 Pro Preview, Gemini 3.8 Flash, and Gemini 3.5 Flash-Lite. Google uses the data from those models to improve its products. "Please don't submit sensitive or personal information" (ai.google.dev/gemini-api/docs/logs-policy). Grounding with Google Search is supported on the free tier for Gemini 3.8 Live, Gemini 3.8 Live Extended Thinking and Gemini 3.1 Flash Live Preview. It is not available on the free tier for Gemini 4 Hydrogen, Gemini 4 Argon or the Gemini 3.x Flash, Flash-Lite and Pro models. On the paid tier, 5,000 search requests a month are free, shared across Gemini 3 and newer models, then $14 per 1,000. Accounts opened after 2026-03-02 cannot spend the $300 Google Cloud welcome credit on the Gemini API.

• The Gemini API Free Tier needs no credit card or billing account; linking a billing account and prepaying moves you to the Paid Tier. (From ai.google.dev/gemini-api/docs/billing, read 2026-10-22.)
• Google's pricing page says free-tier content is used to improve Google's products. (From ai.google.dev/gemini-api/docs/pricing, read 2026-10-04.)
• For developers outside the European Economic Area, Switzerland and the United Kingdom, Google's terms say human reviewers may read free-tier prompts and responses, and Google says not to send sensitive, confidential or personal information on the free tier. (From ai.google.dev/gemini-api/terms, read 2026-10-04.)
• Paid-tier projects created on or after October 26, 2026, store Generate Content API requests and responses by default. Projects created before that date keep their existing Generate Content API storage setting, which is off unless it was turned on in AI Studio. The Interactions API stores requests by default in all projects, including projects created before October 26, 2026. The Paid Tier keeps stored interactions for 55 days and the Free Tier for 1 day. Logging also covers calls made through OpenAI compatibility endpoints. Logs are retained for up to 55 days, configurable to 7, 14 or 28 days. Logs are not used to improve Google's products unless shared. Logging can be toggled independently for each API in AI Studio, and store=false on a request turns it off for that request. (From ai.google.dev/gemini-api/docs/logs-policy, read 2026-11-06.)
• Apps that serve users in the European Economic Area, Switzerland or the United Kingdom may use only the paid tier. (From ai.google.dev/gemini-api/terms, read 2026-10-04.)
• Google's terms require users to be 18 years of age or older to use the Gemini API and Google AI Studio. The API may not be used in a website, app, or other service directed towards or likely to be accessed by individuals under 18. These rules apply to both the free and paid tiers. (From ai.google.dev/gemini-api/terms, read 2026-10-17.)
• Google offers the Gemini API and Google AI Studio, free and paid, in 230 countries and territories, including Taiwan, Brazil, Japan, Singapore, and the United Kingdom. China, Hong Kong, Macao, Russia, Belarus, and Iran are not among them. Google says to developers outside the list: "try the Gemini API in Gemini Enterprise Agent Platform". (From ai.google.dev/gemini-api/docs/available-regions, read 2026-10-17.)
• Google's terms say changes to Gemini API paid prices take effect 30 days after Google posts them unless Google specifies otherwise. Pricing for a new paid service takes effect immediately unless Google specifies otherwise. (From ai.google.dev/gemini-api/terms, read 2026-10-17.)
• Google is moving existing Postpay accounts to Prepay for Gemini API usage. An account that has not switched and added credits by the cutover date in its account notice has its paid service interrupted; other Google Cloud services on the billing account stay on Postpay. (From ai.google.dev/gemini-api/docs/billing, read 2026-10-07.)
• When the Prepay balance reaches $0, every API key in every project on the billing account stops working until credits are added; requests then fail with an HTTP 402 Payment Required error. (From ai.google.dev/gemini-api/docs/billing, read 2026-10-07.)
• The monthly Google Cloud credits from Google AI Pro or Ultra are eligible Google Cloud credits. On the Prepay billing plan, these credits pay for Gemini API usage only after you buy Prepay credits in AI Studio. With a $0 Prepay balance, requests fail even with unused Google Cloud credits, according to a Google staff member on discuss.ai.google.dev. (From discuss.ai.google.dev/t/error-code-402-message-your-prepayment-credits-are-depleted/185524, read 2026-10-10.)
• A paid project does not automatically revert to the free tier, even when its Prepay balance reaches $0. To use the free tier, disable billing on the project, which unlinks it from its billing account. (From ai.google.dev/gemini-api/docs/billing, read 2026-10-10.)
• Disabling billing on a Google Cloud project stops all other paid Google Cloud services in it. Some resources may be deleted and become non-recoverable. Outstanding charges are still due. (From cloud.google.com/billing/docs/how-to/modify-project, read 2026-10-10.)

Labelled speech_api. We read that on 2026-09-01 from https://ai.google.dev/pricing, where it says: “Our 3.1 Flash Text-to-Speech audio model optimized for price-performant, low-latency, controllable speech generation.”. How we use this

Free $0 Forever. Including 2 Analytical Agents, 2 Data Sources, Up to 2 Team Members, 2 Dashboards. Usage Limits 20 requests / day · 50,000 tokens / month

Social media scheduling, publishing and analytics. Free plan at $0/month. The pricing page states no per-plan entitlements: the feature list shown is identical for all four plans.

Stateful AI agents that keep their memory across sessions. The Free plan is $0/month for the Letta app, CLI and web app: up to 3 stateful agents, 5 MiB of storage per memory repository, limited use of Letta Auto, a limited number of LLM requests on rotating free models, and your own model API keys or coding plans. Free is a Personal Plan; API-key access to the Letta API is sold as a Developer Plan, the API Plan at $20/month plus $0.10 per active agent a month and pay-as-you-go model usage. Pro is $20/month for up to 20 stateful agents. The Letta agent harness is open source (Apache 2.0) and runs on your own infrastructure with no Letta account. When we last read the page we cite for this offer, on 2026-11-13, we refused the change we considered recording because it named no figure that had moved, and refusing a change is not a confirmation of the terms above, so we cannot confirm these terms today.

Free plan at $0 per month: 10,000 credits per month and 3 Projects in Studio, with Text to Speech, Speech to Text, Sound Effects, Voice Design, Music, Productions and Image. No commercial licence on the free plan — that starts at Starter, $6 per month for 30,000 credits. Unused credits do not roll over on the free plan, and cancelling a paid plan drops the account back to it. elevenlabs.io/pricing, read 2026-09-19.

Cloudflare AI Search is a managed search service for indexing content and querying it with natural language from a Workers binding, REST API or MCP server. Every account receives 5 million ingestion tokens, 10 GB-month of storage, 1,000 semantic queries, and 1,000 full-text queries each month. The Workers Free plan includes 100 AI Search instances per account, 100,000 files per instance, and 500 pages crawled per day. Billing begins December 1, 2026, and usage before that date is not charged. Beyond the included usage, base ingestion costs $0.75 per million tokens; storage is $2.00 per GB-month; semantic, vector and hybrid queries are $0.75 per 1,000 queries; full-text queries are $0.10 per 1,000 queries. Embedding and reranking usage is included in AI Search usage, while generation and query rewriting are billed separately through Workers AI or the external model provider that runs them. Note — 2026-12-01: Cloudflare AI Search usage is not charged until December 1, 2026, when usage-based billing begins. Every account receives 5 million ingestion tokens, 10 GB-month of storage, 1,000 semantic queries and 1,000 full-text queries each month. Usage beyond these amounts is billed at $0.75 per million ingestion tokens, $2.00 per GB-month of storage, $0.75 per 1,000 semantic queries and $0.10 per 1,000 full-text queries. Cloudflare first set the start for November 1, 2026 and moved it on November 2, 2026. Source ↗