Side-by-side comparison of 10+ serverless free tiers in 2026. Compare AWS Lambda, Google Cloud Functions, Azure Functions, Cloudflare Workers, Deno Deploy, Cloud Run, Val Town, and more — invocations, compute, cold starts, and billing models.
Published 2026-03-31 · Reviewed 2026-11-02 · Figures compiled 2026-03-31, last checked 2026-11-02 · 68 of 103 figures in the tables below come from our records for 1,498 developer tools · 10+ serverless platforms compared
Quick verdict: AWS Lambda gives 1M invocations/month free. Google Cloud Run includes 2M requests, 360K GB-seconds, and 1 GB free egress. Cloudflare Workers includes 100K requests/day, sub-5ms cold starts, and CPU-time billing (you don’t pay for I/O waits). Azure Functions matches Lambda at 1M requests/month on its Consumption plan; its recommended Flex Consumption plan gives 250,000 executions a month.
The critical distinction most developers miss: Cloudflare Workers, Vercel Functions and AWS Lambda Web Functions (since 2026-10-12) bill CPU time, so waiting on I/O is not billed. Lambda Event Functions (AWS's new name for existing Lambda functions), Cloud Run functions and Azure Functions bill elapsed time, waits included. For I/O-heavy workloads this matters more than headline invocation counts.
Side-by-side comparison of the top serverless free tiers.
| Platform | Free Invocations | Free Compute | Memory / CPU | Timeout | Billing Model | Cold Start | Always Free? |
|---|---|---|---|---|---|---|---|
| AWS Lambda CHANGED NOV 7 | 1M / month | 400K GB-sec/mo Source ↗ | 128 MB – 10 GB | 15 min max | Duration (Event Functions); active CPU (Web Functions) | 100–500ms (cold) | Yes (Always Free) |
| |||||||
| Google Cloud Functions No record | 2M / month | 400K GB-sec + 200K GHz-sec Source ↗ | 128 MB – 32 GB | 9 min (1st gen) / 60 min (2nd gen) | Wall-clock time | 200–800ms (cold) | Yes (Always Free) |
| Azure Functions CHANGED NOV 19 Source ↗ | 1M / month (Consumption); 250K (Flex Consumption) | 400K GB-sec/mo (Consumption); 100K (Flex Consumption) | 1.5 GB max (Consumption) | 5 min (default) / 10 min (max) | Wall-clock time | 200–1000ms (cold) | Yes (Always Free) |
| Google Cloud Run CHANGED OCT 30 Source ↗ | 2M / month | 360K GB-sec + 180K vCPU-sec | Up to 32 GB / 8 vCPU | 60 min max | Wall-clock time (request-based) | 100–500ms (cold) | Yes (Always Free) |
| |||||||
| Cloudflare Workers CHANGED SEP 4 Source ↗ | 100K / day (~3M/mo) | 10ms CPU time / invocation | 128 MB RAM | No limit (HTTP) / 15 min (Cron) | CPU time only | <5ms (near-zero) | Yes |
| |||||||
| Vercel Functions CHANGED JUL 20 Source ↗ | 1M function invocations/mo | 4 hrs Active CPU/mo | 2 GB / 1 vCPU (Hobby) | 300s (Hobby) | Active CPU time | <5ms (Edge) / 250ms (Serverless) | Yes (Hobby) |
| |||||||
| Deno Deploy CHANGED OCT 9 Source ↗ | 1M / month | 10 hrs CPU time/mo | 768 MB (default) | No hard limit | CPU time | <100ms (hello world) | No |
| |||||||
| Val Town CHANGED NOV 10 Source ↗ | 100K runs/day | 1 min wall clock / run | Managed (V8 isolates) | 1 min per run | Run-based | <50ms | Yes |
| |||||||
| DBOS Source ↗ | None hosted (open-source library is free) | Self-hosted on your Postgres | Your own servers | No hard limit | DBOS Pro $99/mo; DBOS Cloud via sales | Scale-to-zero (seconds) | Library only |
| Inngest CHANGED JUL 5 Source ↗ | 50K executions/mo | 500K events/mo | Runs on your serverless | Configurable retries | Execution-based | Depends on host | Yes (Hobby) |
| Trigger.dev Source ↗ | $5 free compute/mo | 20 concurrent runs | Managed compute | No hard limit | Compute time-based | Warm pool available | Yes |
The big three clouds' serverless compute, the most mature platforms with the largest ecosystems. Lambda Event Functions, Cloud Run functions and Azure Functions bill elapsed time, waits included. AWS's Lambda Web Functions (since 2026-10-12) bill active CPU time.
| Provider | Free Invocations | Free Compute | Ecosystem | Languages | Best For |
|---|---|---|---|---|---|
| AWS Lambda | 1M / month | 400K GB-sec Source ↗ | API Gateway, DynamoDB, S3, SQS, SNS, EventBridge | Node.js, Python, Go, Java, .NET, Rust, Ruby | AWS-native backends, event-driven architectures |
| Google Cloud Functions No record | 2M / month | 400K GB-sec + 200K GHz-sec Source ↗ | Cloud Run, Firestore, Pub/Sub, Cloud Storage | Node.js, Python, Go, Java, .NET, Ruby, PHP | GCP-native, Firebase integrations |
| Azure Functions Source ↗ | 1M / month | 400K GB-sec | Cosmos DB, Service Bus, Event Grid, Storage | Node.js, Python, Java, .NET, PowerShell | Microsoft/.NET shops, enterprise integrations |
Platforms that bill CPU time rather than elapsed time. Cloudflare Workers runs V8 isolates across Cloudflare's network. Vercel Functions run Node.js, Python and other runtimes in Washington, D.C. by default. Deno Deploy runs Deno apps in Linux microVMs.
| Provider | Free Tier | Cold Start | Runtime | Storage | Best For |
|---|---|---|---|---|---|
| Cloudflare Workers CHANGED SEP 4 Source ↗ | 100K req/day, 10ms CPU/req | <5ms | V8 isolates (JS/TS/Wasm) | KV: 1 GB, D1: 5 GB, R2: 10 GB | APIs, edge logic, full-stack apps |
| Vercel Functions Source ↗ | 1M invocations/mo, 4 hrs Active CPU | <5ms | Node.js, Python, Go, Ruby, Bun, Rust, Wasm | Blob: 1 GB; Redis via Marketplace (Vercel KV ended) | Next.js middleware, edge API routes |
| Deno Deploy CHANGED OCT 9 Source ↗ | 1M req/mo, 10 hrs CPU, 20 GiB egress | <100ms (hello world) | Deno runtime (JS/TS, Web APIs) | KV: 1 GB (1M reads/mo) | Deno/Fresh apps, Web API-native code |
Container-based serverless that runs any Docker image — more flexibility than FaaS, with scale-to-zero on the free tier.
| Provider | Free Tier | Container Support | Concurrency | Egress | Best For |
|---|---|---|---|---|---|
| Google Cloud Run CHANGED OCT 30 Source ↗ | 2M req/mo, 360K GB-sec, 180K vCPU-sec | Any Docker image | Up to 1000 concurrent per instance | 1 GB North America free | Containerized APIs, microservices |
| AWS App Runner | Closed to new customers since April 30, 2026. Existing customers only. Source ↗ | Any Docker image or source code | Configurable | Standard AWS egress | For existing App Runner customers. New customers use ECS Express Mode. |
Purpose-built serverless platforms for specific workloads — scripting, durable workflows, background jobs, and orchestration.
| Provider | Free Tier | Specialty | Durability | Best For |
|---|---|---|---|---|
| Val Town CHANGED NOV 10 Source ↗ | 100K runs/day, unlimited public vals | Script-as-a-service, shareable functions | SQLite storage, blob storage | Existing vals until January 10, 2027; no new signups |
| DBOS Source ↗ | Open-source library free; DBOS Cloud via sales, Pro $99/mo | Durable workflow orchestration | Built-in (transactional, exactly-once) | Transactional workflows, reliable background processing |
| Inngest CHANGED JUL 5 Source ↗ | 50K executions/mo, 500K events | Event-driven workflow orchestration | Built-in retries, step functions | Background jobs, scheduled tasks, event-driven workflows |
| Trigger.dev Source ↗ | $5 free compute/mo, 20 concurrent | Background jobs with managed compute | Durable execution, retries | Long-running background jobs, scheduled tasks |
| Cloudflare Durable Objects Source ↗ | 100K requests/day (on Workers free) | Stateful serverless compute | Strongly consistent, WebSocket hibernation | Real-time collaboration, stateful edge computing |
| ||||
This is the most important and least understood difference between serverless platforms. It can change your effective cost by 10-50x.
You pay for the whole run, waits included. A 500 ms call that waits 480 ms on a database is billed for 500 ms.
You pay for CPU time actually used; waiting on I/O is not billed. The same call that uses 20 ms of CPU is billed for 20 ms.
Lambda for event-driven microservices in the AWS ecosystem. Cloud Run for containerized APIs with the most generous free tier (2M req/month). Both handle production-scale traffic. Cloud comparison →
100K requests/day with sub-5ms cold starts, globally distributed. CPU-time billing makes I/O-heavy workloads dramatically cheaper. Integrated KV, D1, R2, and Durable Objects storage. View Workers details →
Cloudflare Workers: write a function in the Cloudflare dashboard and get a workers.dev URL. The Free plan gives 100,000 requests a day.
Vercel deprecated Edge Functions for new projects; use Vercel Functions on the Node.js runtime with Fluid compute. Hobby includes 1M invocations and 4 hours of Active CPU a month, for non-commercial use.
On AWS? Use Lambda. On GCP? Use Cloud Functions or Cloud Run. On Azure? Use Azure Functions. The free tiers are similar enough that ecosystem integration and team expertise matter more than marginal differences in invocation limits.
Inngest: event-driven workflow orchestration with 50K executions/month free. Trigger.dev: managed compute for long-running background jobs with $5 free compute/month. Both handle retries, scheduling, and durable execution. Hosting alternatives →
DBOS Transact is open source and free to self-host on Postgres, for TypeScript, Python, Go, Java and Kotlin. There is no free hosted plan: DBOS Pro is $99 a month and DBOS Cloud is sold through sales.
Strongly consistent state at the edge with WebSocket hibernation. 100K requests/day on the free Workers plan. Unique capability — no other edge platform offers per-object consistency guarantees.
Serverless free tiers have several non-obvious costs that can surprise you. Here’s what to watch for.
A Lambda function URL serves HTTPS at no charge beyond Lambda's own. API Gateway is for its extra features (request validation, authorizers, Cognito, OAuth 2.0) at $3.50 per million REST API requests or $1.00 per million HTTP API requests.
Most serverless platforms charge for outbound data transfer beyond free tiers. AWS: 100 GB/month free on an account of any age, aggregated across all AWS services, then $0.09/GB. GCP: 1 GB/month free (North America). Azure: 100 GB/month free. Cloudflare Workers: Unlimited egress (no bandwidth charges). Deno Deploy: 20 GiB/month free. For bandwidth-heavy workloads, Workers’ unlimited egress is a major differentiator.
AWS Lambda Provisioned Concurrency eliminates cold starts but costs $0.015/GB-hour — not free tier eligible. Azure Functions Premium keeps instances warm but starts at ~$0.173/vCPU-hour. Cloudflare Workers avoids this with V8 isolates, with cold starts under 5 ms and no warm-up configuration. Deno Deploy pre-provisions Linux microVMs, and Deno puts its cold starts under 100 ms for a hello-world app.
On 2025-09-09 Vercel moved Pro to $20 of monthly usage credit. Hobby is unchanged and has no credits: it includes 1M function invocations, 4 hours of Active CPU and 360 GB-hours of Provisioned Memory a month, and Hobby accounts cannot buy more. Vercel details →
Cloudflare Workers free tier limits each invocation to 10ms CPU time. This is CPU time, not wall-clock time, so I/O waits don’t count — but CPU-intensive tasks (heavy JSON parsing, image processing, crypto) can hit this limit. Paid Workers ($5/month) increases to 30 seconds CPU time per invocation. For most API workloads, 10ms CPU is sufficient.
Recent serverless-related pricing changes from our tracker. See the full timeline for all 569 tracked changes.
| Date | Provider | Change | Impact |
|---|---|---|---|
| effective Nov 20, 2026 | Cloudflare Workers AI | Workers AI keeps running. Cloudflare removes 24 Workers AI models on November 20, 2026: they can be called until then, and from that date requests to them fail with HTTP 410 Gone (error code 5028). Language models: @cf/moonshotai/kimi-k2.6, @cf/moonshotai/kimi-k2.7-code, @cf/zai-org/glm-5.2, @cf/zai-org/glm-5.3-flash, @cf/deepseek-ai/deepseek-v4-flash-0731, @cf/deepseek-ai/deepseek-v4-pro-0813, @cf/deepseek-ai/deepseek-r1-distill-qwen-32b, @cf/nvidia/nemotron-3-120b-a12b, @cf/qwen/qwq-32b, @cf/qwen/qwen3-30b-a3b-fp8, @cf/meta/llama-3.1-8b-instruct-fp8, @cf/meta/llama-3.2-1b-instruct, @cf/meta/llama-3.2-3b-instruct, @cf/meta-llama/llama-2-7b-chat-hf-lora, @cf/google/gemma-2b-it-lora and @cf/google/gemma-7b-it-lora; Cloudflare recommends @cf/deepseek-ai/deepseek-v4.1-flash-1017 or @cf/zai-org/glm-5.3 for frontier model needs and @cf/google/gemma-4-26b-a4b-it or @cf/qwen/qwen3.8-27b for smaller ones. Other models, each with Cloudflare's replacement: @cf/myshell-ai/melotts (@cf/deepgram/aura-2-en); @cf/stabilityai/stable-diffusion-xl-base-1.0, @cf/runwayml/stable-diffusion-v1-5-inpainting, @cf/bytedance/stable-diffusion-xl-lightning and @cf/lykon/dreamshaper-8-lcm (@cf/black-forest-labs/flux-2-klein-9b); @cf/openai/whisper-tiny-en (@cf/deepgram/nova-3); @cf/llava-hf/llava-1.5-7b-hf (@cf/moondream/moondream3.1-9B-A2B); @cf/google/embeddinggemma-300m (@cf/google/embeddinggemma-2, not yet available). Source ↗ | MEDIUM |
| effective Nov 19, 2026 | Azure | Azure OpenAI retires o1 (version 2024-12-17), o1-pro (2025-03-19), o3 (2025-04-16), o3-deep-research (2025-06-26), o3-mini (2025-01-31), o3-pro (2025-06-10) and o4-mini (2025-04-16) on November 19, 2026. These models are deprecated, so new customers cannot deploy them. After that date a request to a retired model returns HTTP 410 Gone. Microsoft names gpt-5.6-sol as the replacement for o1, o1-pro, o3, o3-deep-research and o3-pro, and gpt-5.6-terra for o3-mini and o4-mini. Microsoft upgrades Global Standard, Data Zone Standard and Standard deployments to the named replacement where an upgrade is supported, and does not upgrade Provisioned deployments. Microsoft Foundry also retires Llama-4-Maverick-17B-128E-Instruct-FP8 that day, with no replacement named. Source ↗ | MEDIUM |
| effective Nov 10, 2026 | Val Town | Val Town can be used until January 10, 2027, when all its services end. Val Town announced on 2026-11-10 that it is winding down: new signups and plan upgrades ended that day, and Val Town becomes read-only on December 10, 2026, except SQLite and blob storage. Billing is paused, and existing subscriptions continue free until the end. Source ↗ | HIGH |
| effective Nov 7, 2026 | AWS | Amazon OpenSearch Service charges an Extended Support surcharge equal to the instance price on every domain still running Elasticsearch 1.5, 2.3, 5.1, 5.3, 5.5, 6.0, 6.2, 6.3, 6.4, 6.5, 6.7, 7.1, 7.4, 7.7 or 7.8, or OpenSearch 1.0, 1.1, 1.2, 2.3, 2.5, 2.7 or 2.9 from November 7, 2026. The surcharge doubles the instance cost. Storage costs are not affected. Before November 7, 2026 these OpenSearch Service domains paid $0.0065 per Normalized Instance Hour (NIH) in US East (N. Virginia). Amazon OpenSearch Service's example: an m7g.medium.search instance at $0.068 an hour on demand costs $1.632 a day. The surcharge adds another $1.632 a day. The surcharge applies automatically and stops when the domain is upgraded to a supported version. This rate applies until November 7, 2027. Elasticsearch 5.6 domains still pay $0.0065 per NIH. Elasticsearch 6.8, 7.9, 7.10 and OpenSearch 1.3, 2.11, 2.13, 2.15, 2.17 and 2.19 pay no surcharge before November 7, 2027. Source ↗ | MEDIUM |
| effective Oct 30, 2026 | Google Cloud Run | Google decommissioned the Node.js 20 runtime of Cloud Functions for Firebase and Cloud Run functions on October 30, 2026. Since then, functions on Node.js 20 cannot be created or updated. The Firebase CLI refuses to deploy them with the error "Runtime Node.js 20 was decommissioned on 2026-10-30". Functions already deployed on Node.js 20 can keep running, but Google says functions that stay on a decommissioned runtime may be disabled. To deploy again, move the functions to Node.js 22 or Node.js 24. Source ↗ | MEDIUM |
| effective Oct 9, 2026 | Deno Deploy | Deno says Deno Deploy will be sunset in April 2027 and gives no day. Existing applications keep running until then. Deno announced on October 9, 2026 that its team is joining Cloudflare. Deno will help paying customers move to Cloudflare Workers. Source ↗ | HIGH |
| effective Oct 7, 2026 | AWS | AWS Proton reaches end of support on 2026-10-07, when its console and resources stop being accessible. AWS: "Your deployed CloudFormation stacks and the resources they manage will remain intact and continue to function." New customers could not sign up after 2025-10-07. Source ↗ | MEDIUM |
| effective Sep 30, 2026 | AWS | AWS App Mesh reached end of support on 2026-09-30. AWS: "After September 30, 2026, you will no longer be able to access the AWS App Mesh console or AWS App Mesh resources." AWS points App Mesh users to Amazon ECS Service Connect. Source ↗ | MEDIUM |
| discovered Sep 5, 2026 · effective date unknown | Cloudflare Queues | The free tier now offers 10,000 operations/day, and the paid tier offers 1,000,000 operations/month. Message retention on the free tier is 24 hours (non-configurable), while the paid tier offers 4 days default, configurable up to 14 days. The pricing model is now based on operations (64KB chunks of data written, read, or deleted) with additional costs per million operations beyond the included allowance. Unconfirmed. We could not check this change against an archived copy of Cloudflare Queues's page, so it does not count toward the rating. Source ↗ | HIGH |
| effective Sep 4, 2026 | Cloudflare Workers | On September 4, 2026, Cloudflare removed the compressed Worker size limits of 3 MB on Workers Free and 10 MB on Workers Paid. A Worker on either plan can now be up to 64 MiB, counted on the uncompressed bundle. Source ↗ | MEDIUM |
| discovered Aug 28, 2026 · effective date unknown | Deno Deploy | Deno Deploy cut its Free plan's CPU time from 15 to 10 hours a month, memory time from 350 GB-hours to 150 GiB-hours, custom domains from 50 to 5, apps from 20 to 10 and team members from 5 to 3; 60 deployments an hour became 15 builds an hour, one at a time. KV read units went up from 450,000 to 1,000,000 a month and write units from 300,000 to 500,000. The pricing page changed between a Common Crawl capture of 2026-08-07 and our read on 2026-08-28. Source ↗ | HIGH |
| effective Jul 28, 2026 | Cloudflare Workers AI | 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 ↗ | HIGH |
These are announced changes that have not taken effect yet, listed soonest first.
| Date | Vendor | Change | Impact |
|---|---|---|---|
| effective Dec 7, 2026 | AWS Infrastructure Composer | The standalone AWS Infrastructure Composer console shuts down on December 7, 2026. New customers have not been able to sign up since October 29, 2026. Existing customers can use the console until December 7, 2026. The visual editor remains in the AWS Toolkit for Visual Studio Code. CloudFormation templates and stacks continue to work normally. Source ↗ | MEDIUM |
| effective Dec 15, 2026 | Azure | Azure OpenAI retires whisper, tts and tts-hd (version 001) on December 15, 2026. After that date every request to them returns HTTP 410 Gone. Microsoft names no replacement. OpenAI's own API retires whisper-1 later, in February 2027, and tts-1 and tts-1-hd later, in January 2027. Those are OpenAI's dates, and Azure OpenAI's are Microsoft's. Source ↗ | MEDIUM |
| effective Dec 15, 2026 | Azure | Azure OpenAI retires gpt-realtime-mini version 2025-12-15 on December 15, 2026. gpt-realtime-mini version 2025-10-06 retires later on Azure OpenAI, in April 2027, and is not covered by this date. Microsoft names no replacement. After that date requests to version 2025-12-15 return HTTP 410 Gone. OpenAI's own API retires gpt-realtime-mini later, on January 20, and that is OpenAI's date, not Azure's. Source ↗ | MEDIUM |
| effective Jan 1, 2027 | Google Cloud SQL for MySQL | From January 1, 2027, Google charges extended support on every Cloud SQL for MySQL 8.0 instance, on top of the regular instance price. Cloud SQL enrolls them automatically. MySQL 8.4 or later instances are not affected. In Iowa (us-central1), Cloud SQL extended support for MySQL 8.0 costs $0.07 per vCPU-hour in years 1 and 2 and twice that in year 3. High-availability instances pay twice the per-vCPU rate. On shared-core instances the extended support charge is $0.018 an hour for db-f1-micro and $0.035 for db-g1-small. Committed use discounts do not apply. The charge stops when the instance is upgraded. On July 1, 2029, Cloud SQL upgrades remaining MySQL 8.0 instances to the default major version automatically. Source ↗ | MEDIUM |
| effective Jan 1, 2027 | AWS | AWS discontinues its Developer Support and Business Support plans on January 1, 2027. Accounts on either plan must opt in to Business Support+ before then. A Developer Support account that does not opt in is downgraded to Basic Support. Basic Support gives access to the Resource Center, Service Health Dashboard, Product FAQs and Discussion Forums at no charge. Business Support accounts are not moved to Business Support+ automatically. Business Support+ costs the greater of $29 a month per account or 9% of monthly AWS charges, falling in steps to 3% on larger bills. Neither plan has accepted new subscriptions since December 2, 2025. AWS emails each account 30 days before its plan changes. Enterprise On-Ramp also ends on January 1, 2027, and its customers are moved to Enterprise Support automatically. The three plans stay available in the AWS GovCloud (US) Region. Source ↗ | MEDIUM |
| effective Jan 8, 2027 | AWS | Amazon Bedrock removes Claude Opus 4.1 on January 8, 2027. The date covers us-east-1, us-east-2 and us-west-2. Requests to it on Bedrock fail after that date. It has been Legacy on Bedrock since July 8, 2026. New customers cannot use it. Accounts that leave it unused for 15 days may lose access. Since October 8, 2026 it has been in public extended access. Bedrock says to expect higher prices set by Anthropic. Anthropic's own Claude API retired Claude Opus 4.1 on August 5, 2026. Bedrock's dates apply only to Bedrock. Bedrock lists Claude Opus 4.8, Claude Opus 5 and Claude Opus 5.5. Source ↗ | LOW |
| effective Jan 10, 2027 | Val Town | Val Town can be used until January 10, 2027, when all its services end. Val Town announced on 2026-11-10 that it is winding down: new signups and plan upgrades ended that day, and Val Town becomes read-only on December 10, 2026, except SQLite and blob storage. Billing is paused, and existing subscriptions continue free until the end. Source ↗ | HIGH |
| effective Jan 14, 2027 | Azure | Azure OpenAI retires gpt-4o version 2024-05-13 on January 14, 2027. gpt-4o versions 2024-08-06 and 2024-11-20 and gpt-4o-mini version 2024-07-18 retire later on Azure OpenAI, on April 14, 2027, in all active deployments including Global Standard. Microsoft upgrades Global Standard and Data Zone Standard deployments of version 2024-05-13 to gpt-5.1 version 2025-11-13 that day. Provisioned deployments of version 2024-05-13 are not upgraded. Create a gpt-5.1 deployment and move traffic before the date. Microsoft does not promise automatic upgrades for Global Batch deployments of version 2024-05-13. After that date, requests to version 2024-05-13 return HTTP 410 Gone. Source ↗ | MEDIUM |
| effective Feb 1, 2027 | AWS | Amazon ElastiCache charges an Extended Support premium on every cache still running Redis OSS 6 starting February 1, 2027. Standard support for Redis OSS 6 ends January 31, 2027. Caches move to Extended Support automatically. The premium is 80% of the on-demand node price in years 1 and 2 and 160% in year 3. Reserved node discounts do not apply to the premium. AWS's example: a cache.m5.large node in US East (Ohio) costs $0.156 an hour plus $0.1248 an hour. The premium stops when the cache is upgraded to Valkey or Redis OSS 7 or later, or deleted. Source ↗ | MEDIUM |
| effective Feb 1, 2027 | Google Cloud SQL for PostgreSQL | From February 1, 2027, Google charges extended support on every Cloud SQL for PostgreSQL 14 instance, on top of the regular instance price. Cloud SQL enrolls them automatically. Databases on PostgreSQL 15 or later are not affected. In Iowa (us-central1), Cloud SQL extended support costs $0.07 per vCPU-hour in years 1 and 2 ($0.14 for HA vCPUs), $0.14 per vCPU-hour in year 3, and $0.018 an hour on a db-f1-micro instance. Committed use discounts do not apply. Read replicas pay the same rate. There is no setting to turn it off. The charge stops when the instance is upgraded. On February 1, 2030, Cloud SQL upgrades remaining PostgreSQL 14 instances to the default major version automatically. Source ↗ | MEDIUM |
| effective Feb 1, 2027 | Azure | Azure Database for PostgreSQL flexible server 14 leaves standard support on January 31, 2027. From February 1, 2027 every server still on PostgreSQL 14 is enrolled in Extended Support automatically. Servers on PostgreSQL 15 or later are not affected. There is no opt-out. Charges stop only when the server is upgraded to a supported version. Azure charges $0.07 per vCore-hour in East US in years 1 and 2. Year 3 is charged at twice the year 1 and 2 rate. Only running servers are charged. Stopped servers are not charged. Extended Support for PostgreSQL 14 ends November 11, 2029. Source ↗ | MEDIUM |
| effective Feb 1, 2027 | Azure | Azure charges Extended Support on every Azure Database for MySQL flexible server 8.0 server starting February 1, 2027, after standard support ends on January 31, 2027. Azure enrolls them automatically. Servers on MySQL 8.4 or later are not affected. Azure charges $0.07 per vCore-hour in East US in years 1 and 2, and twice that in year 3. Stopped servers are not charged for Extended Support. There is no opt-out: the charge stops when the server is upgraded to a supported major version. Extended Support for MySQL 8.0 ends May 31, 2029. Source ↗ | MEDIUM |
| effective Feb 22, 2027 | Azure Deployment Environments | Azure Deployment Environments retires on February 22, 2027. Creating and deploying environments and other write operations are expected to be blocked on that date. Inventory, read, log and delete stay available for a limited cleanup period. Microsoft Dev Box retires later, in September 2028. Source ↗ | MEDIUM |
| effective Mar 1, 2027 | AWS | On March 1, 2027, AWS charges for RDS Extended Support on Amazon RDS for PostgreSQL 14 and Aurora PostgreSQL 14 databases that are enrolled and still on version 14. Standard support for PostgreSQL 14 on RDS and Aurora ends February 28, 2027. Databases on PostgreSQL 15 or later are not affected. Each database's RDS Extended Support setting decides what happens: an enrolled PostgreSQL 14 database is charged, and a PostgreSQL 14 database with it disabled is upgraded to a supported version on or shortly after February 28, 2027. The AWS CLI, the RDS API and CloudFormation enable it by default, and the console does not select it for new databases. The charge stops when the database is upgraded to a version under standard support or deleted. For earlier versions AWS charges $0.10 per vCPU-hour in years 1 and 2 in US East (Ohio), and Aurora Serverless v2 on PostgreSQL 14 is $0.085 per ACU-hour in years 1 and 2. On the same date RDS for PostgreSQL 12 moves from $0.100 to $0.200 per vCPU-hour in US East (Ohio). Source ↗ | MEDIUM |
| effective Mar 30, 2027 | AWS | Amazon Bedrock removes 26 models on March 30, 2027. DeepSeek: DeepSeek-R1, DeepSeek-V3.1. Google: Gemma 3 4B IT, Gemma 3 12B IT, Gemma 3 27B PT. Meta: Llama 3 8B Instruct, Llama 3 70B Instruct, Llama 3.1 8B Instruct, Llama 3.1 70B Instruct, Llama 3.3 70B Instruct, Llama 4 Maverick 17B Instruct, Llama 4 Scout 17B Instruct. MiniMax: MiniMax M2, MiniMax M2.1. Mistral AI: Devstral 2 123B, Mistral 7B Instruct, Mixtral 8x7B Instruct, Voxtral Mini 3B 2507, Voxtral Small 24B 2507. Moonshot AI: Kimi K2 Thinking. NVIDIA: NVIDIA Nemotron Nano 9B v2, NVIDIA Nemotron Nano 12B v2 VL BF16. Qwen: Qwen3-Coder-30B-A3B-Instruct, Qwen3 Coder 480B A35B Instruct. Z.AI: GLM 4.7, GLM 4.7 Flash. They have been Legacy on Bedrock since September 30, 2026. New customers cannot use them. Accounts that leave one unused for 15 days may lose access. No new Provisioned Throughput can be created. Bedrock moves them to public extended access before the end date, with higher prices set by the provider. It has not announced when. After March 30, 2027 requests to them on Bedrock fail. Bedrock does not migrate applications automatically. The date covers the Regions Bedrock lists for each model, us-west-2 among them for all 26. A model can stay available in other Regions. Bedrock names no replacement. Its model cards list newer models: DeepSeek V3.2, GLM 5, Gemma 4, Kimi K3, MiniMax M2.5 and Mistral Large 3. Source ↗ | MEDIUM |
| effective Mar 31, 2027 | Azure | Microsoft retires two tiers of Azure Cache for Redis on March 31, 2027: Enterprise and Enterprise Flash. The Basic, Standard and Premium tiers retire later, on September 30, 2028. From April 1, 2027 every remaining Enterprise cache and every remaining Enterprise Flash cache is disabled. Creating new Enterprise or Enterprise Flash caches has been blocked since April 1, 2026. Reservations for both tiers are supported until March 30, 2027. Microsoft's replacement is Azure Managed Redis. Applications must switch to the new cache's hostname and access credentials. Azure Managed Redis is clustered by default, so applications should check that their client library works with a clustered cache. New customers cannot create Basic, Standard or Premium caches either. Since April 1, 2026, in Azure's public cloud, only a tenant that had any Azure Cache for Redis instance before that date can create them. Source ↗ | MEDIUM |
| effective Apr 4, 2027 | Google Cloud Run | Google decommissions the Python 3.10 runtime of Cloud Functions for Firebase and Cloud Run functions on April 4, 2027. From that date, functions on Python 3.10 cannot be created or updated. Google says functions that stay on a decommissioned runtime may be disabled. Python 3.10 has been deprecated since October 4, 2026. Until April 4, 2027, functions on Python 3.10 can still be created and updated. Move them to Python 3.11 or a later Python runtime before then. The Node.js 20 runtime of these products was already decommissioned on October 30, 2026. Node.js 20 functions can no longer be deployed. The Firebase CLI refuses with the error "Runtime Node.js 20 was decommissioned on 2026-10-30". Source ↗ | MEDIUM |
| effective Apr 8, 2027 | AWS | Amazon Bedrock removes Claude Sonnet 4.5 on April 8, 2027. The date covers its commercial AWS Regions. In the AWS GovCloud (US) Regions it is later in April 2027. Requests to it on Bedrock fail after that date. Anthropic's own Claude API retires the model at the end of November 2026. Bedrock says only its own dates apply to Bedrock usage. It has been Legacy on Bedrock since October 8, 2026. New customers cannot use it. Accounts that leave it unused for 15 days may lose access. From January 2027 it is in public extended access. Bedrock says to expect higher prices set by Anthropic. Bedrock lists Claude Sonnet 5 and Claude Sonnet 5.5. Source ↗ | MEDIUM |
| effective Apr 14, 2027 | Azure | Azure OpenAI retires gpt-4o versions 2024-08-06 and 2024-11-20, gpt-4o-mini version 2024-07-18, and gpt-4.1, gpt-4.1-mini and gpt-4.1-nano versions 2025-04-14 on April 14, 2027. Regional Standard deployments that already retired stay retired. Microsoft names gpt-5.1 as the replacement for gpt-4o. No replacement is named for the others. Fine-tuned deployments of these models have later dates on a separate Microsoft schedule. OpenAI's own API shuts down gpt-4o, gpt-4o-mini, gpt-4.1 and gpt-4.1-mini earlier, in January 2027. That is OpenAI's date, not Azure's. Source ↗ | MEDIUM |
| effective Jun 2, 2027 | Azure Fluid Relay | Azure Fluid Relay retires on June 2, 2027. It can be used until June 1, 2027. The Fluid Framework remains open source. Replace it with a self-hosted Fluid service you run yourself. Source ↗ | MEDIUM |
| effective Jun 16, 2027 | Azure | Azure OpenAI retires gpt-image-1.5 version 2025-12-16 on June 16, 2027. Microsoft names no replacement. After that date requests to it return HTTP 410 Gone. OpenAI's own API retires gpt-image-1.5 earlier, on December 1, and that is OpenAI's date, not Azure's. Source ↗ | MEDIUM |
| effective Jun 30, 2027 | Google Cloud Deployment Manager | Google Cloud Deployment Manager is turned down after June 30, 2027. Support ended April 1, 2026. Since June 30, 2026, new users cannot enable the Deployment Manager API or create a first deployment. Existing deployments can use Deployment Manager until June 30, 2027. After that date, Deployment Manager can no longer be used. Resources it created continue to run and are managed with standard Google Cloud tools. Google recommends Infrastructure Manager. Source ↗ | MEDIUM |
| effective Aug 31, 2027 | AWS | AWS Lambda blocks creating new functions on its deprecated runtimes from July 29, 2027. It blocks updating the code and configuration of existing functions from August 31, 2027. AWS is delaying these blocks beyond its usual 30 and 60 days after deprecation. Functions on .NET 8, Python 3.10 and the other deprecated runtimes can still be created and updated until those dates through the AWS CLI, AWS SAM or CloudFormation, not the Lambda console. After both dates, existing functions can still be invoked and can still be switched to a supported runtime. Deprecated runtimes include Node.js 16, 18 and 20; Python 3.8, 3.9 and 3.10; Ruby 3.2; .NET 6 and 8; and provided.al2. Python 3.10 was deprecated on October 31, 2026 and .NET 8 on November 10, 2026. These runtimes are deprecated next and get the same two dates: Ruby 3.3 on March 31, 2027, Node.js 22 on May 30, 2027, and Python 3.11 and Java 8, 11 and 17 on Amazon Linux 2 on June 30, 2027. Older runtimes are the exception: functions on Node.js 14, Python 3.7, Ruby 2.7, Go 1.x, Java 8 on Amazon Linux and provided can no longer be created since 2024 and can no longer be updated from August 31, 2027. Source ↗ | MEDIUM |
| effective Sep 14, 2027 | Google Cloud Video Intelligence API | Google Cloud Video Intelligence API shuts down on September 14, 2027. It has been deprecated and unsupported since September 14, 2026, and can be used until its shutdown. Google recommends migrating to its Gemini family of models. Source ↗ | MEDIUM |
| effective Sep 29, 2027 | Functions on Azure Container Apps V1 | Functions on Azure Container Apps V1 retires on September 29, 2027. After that date V1 function apps stop running and stop processing requests and triggers. Migrate to Functions on Azure Container Apps V2, a native Container App with kind=functionapp, using the same container image and no function code changes in the standard path. Azure Functions and Azure Container Apps keep running. Source ↗ | MEDIUM |
| effective Nov 7, 2027 | AWS | After November 7, 2027, Amazon OpenSearch Service isolates every domain still running Elasticsearch 1.5, 2.3, 5.1, 5.3, 5.5, 6.0, 6.2, 6.3, 6.4, 6.5, 6.7, 7.1, 7.4, 7.7 or 7.8, or OpenSearch 1.0, 1.1, 1.2, 2.3, 2.5, 2.7 or 2.9. Isolated domains can no longer be accessed. AWS keeps their data for at least 30 days. Data must be restored into a new domain on a supported version. New domains cannot be created on these versions after that date. Since November 7, 2026, these domains have paid an Extended Support surcharge equal to their instance price, doubling the cost. Elasticsearch 1.5 and 2.3 domains cannot be upgraded in place. Upgrading before November 7, 2027 stops the surcharge and avoids isolation. Source ↗ | MEDIUM |
| effective Sep 30, 2028 | Azure Communication Services | Azure Communication Services' standalone services retire on September 30, 2028. Retired services include Email, SMS, WhatsApp messaging, Chat, Rooms, number management, Job Router, Direct Routing and the UI Library SDKs. Customers without an ACS resource created before October 23, 2026 cannot acquire phone numbers for SMS or PSTN calling. Voice and video calling stay available after 2028 only with Microsoft Teams integrations, on new SDK major versions. Source ↗ | HIGH |
| effective Sep 30, 2028 | Azure | Microsoft retires the Basic, Standard and Premium tiers of Azure Cache for Redis on September 30, 2028. From October 1, 2028 every remaining Basic, Standard or Premium cache is disabled. Since April 1, 2026 new customers in Azure's public cloud cannot create caches in these tiers. A tenant that had any Azure Cache for Redis instance before April 1, 2026 can keep creating them until the retirement. Reservations are supported until September 30, 2028. Microsoft's replacement is Azure Managed Redis. Applications must switch to the new cache's hostname and access key. Azure Managed Redis is clustered by default, so the client library must be set up for a clustered cache. Source ↗ | MEDIUM |
A source check reads each cited page for that service's name and a price. The figures in the tables above are from our own records, except for the services whose entry below dates them to a read of the vendor's own page.
We name no most generous free tier here, because generosity is not one quantity: the columns above measure different things and a tier that leads one trails another.
Each provider structures free tier limits differently — some cap storage, others cap requests or compute hours. Our comparison table provides exact numbers for each provider.