Complete OpenAI Assistants API migration guide. Feature mapping to Responses API, migration complexity assessment, decision framework (stay vs switch vs go agnostic), 15+ alternatives compared, cost analysis. Shutdown August 26, 2026.

OpenAI Assistants API Migration Guide 2026

Published 2026-04-03 · Not yet reviewed · Figures in the tables below come from our records for 1,498 developer tools · 19 OpenAI pricing changes tracked

Shut down
August 26, 2026
OpenAI stability: VOLATILE
Shut down
August 26, 2026
12
Alternatives Compared
3
Migration Paths
8
Feature Mappings

The situation: OpenAI deprecated the Assistants API on August 26, 2025. The v1 beta access ended December 18, 2024. The full shutdown is August 26, 2026 — after which all Assistants, Threads, Runs, and Messages endpoints stop functioning. OpenAI’s official migration guide has significant gaps: references non-existent UI elements, no automated tooling, and no code examples for complex patterns.

Key insight: The Responses API is a capable replacement with new features (MCP support, deep research, web search, computer use), but it has breaking changes that the official guide undersells: no programmatic Prompt creation, .docx upload regression, shifted state management burden, and 30-day response TTL. Community sentiment reflects frustration with trust erosion and naming confusion (Chat → Prompts → Assistants → Responses).

Our data: OpenAI’s stability rating is volatile based on 19 tracked pricing changes. This guide covers three paths: migrate within OpenAI, switch to another provider, or go provider-agnostic to avoid future deprecation cycles.

Jump to section

  1. Timeline & Status
  2. Feature Migration Map
  3. Migration Complexity Assessment
  4. Decision Framework: Migrate vs. Leave
  5. Alternatives: APIs, Frameworks & Bridges
  6. Token Cost Comparison
  7. OpenAI Pricing Change Timeline
  8. Methodology

1. Timeline & Status

Key dates for the Assistants API deprecation and shutdown. Plan your migration timeline around these milestones.

Aug 26, 2025
Deprecation announced. OpenAI announced Assistants API deprecation alongside Responses API launch. One-year migration window begins.
Dec 18, 2024
v1 beta access ended. Assistants API v1 beta endpoints stopped accepting new requests. All users must be on v2.
Aug 26, 2026
Full shutdown. All Assistants, Threads, Runs, and Messages endpoints cease functioning. No grace period announced. Microsoft retired the Azure OpenAI Assistants API on August 26, 2026 too and directs Azure agents to Microsoft Foundry Agent Service.
What happens after shutdown: Since the shutdown on August 26, 2026, Assistants API calls no longer work, including the call that retrieves thread messages; OpenAI says to migrate history from messages your application stored.

2. Feature Migration Map

Every Assistants API feature mapped to its Responses API equivalent, with migration complexity and gotchas. 8 features mapped.

Assistants API Responses API Replacement Complexity Notes & Gotchas
Assistants (persistent config) Prompts (dashboard-only, NOT API-creatable) Medium Breaking: no programmatic creation. Must use dashboard or inline instructions.
Threads (conversation state) Conversations API (or self-managed context) High State management shifts to developer. 30-day response TTL.
Runs / Run Steps Responses / Items Low Streaming-first. Simpler lifecycle, no polling required.
Code Interpreter Code interpreter tool Low Direct equivalent. Same capability, different API shape.
File Search (vector stores) File search tool Medium .docx upload regression reported. PDF and other formats work.
Function Calling Function calling (compatible) Low Same concept. Plus new MCP server support.
Annotations / Citations Citations in responses Low Format changed but concept preserved.
N/A (new) MCP support, deep research, web search, computer use — New capabilities not available in Assistants API.
Key breaking changes the official guide undersells: (1) No programmatic Prompt creation — Prompts (the replacement for Assistants) can only be created in the dashboard, not via API. Dynamic assistant creation patterns break. (2) .docx upload regression — file search has a reported regression with Word documents. (3) State management shifted to developer — Threads managed state for you; now you manage conversation context yourself or use the new Conversations API. (4) 30-day response TTL — Responses API data expires after 30 days by default.
New capabilities worth noting: The Responses API adds features the Assistants API never had: MCP server support (connect to external tools), deep research (multi-step information gathering), web search (real-time information), and computer use (browser-based task execution). If you were building workarounds for these in the Assistants API, migration may actually simplify your code.

3. Migration Complexity Assessment

Not all migrations are equal. Your complexity depends on which Assistants API features you used and how deeply.

✅ Low Complexity Patterns

⚠️ High Complexity Patterns

4. Decision Framework: Migrate vs. Leave

Three paths depending on your priorities. Each has different trade-offs in migration effort, cost, and future-proofing.

🔄 Path 1: Stay with OpenAI — Migrate to Responses API

Direct replacement with feature parity plus new capabilities. Threads → Conversations API. Assistants → Prompts + system instructions. Code Interpreter and File Search tools carry over. New: MCP support, deep research, web search, computer use.

Choose this when: You’re deeply invested in OpenAI-specific features (web search, code interpreter), need the lowest migration effort, or have production apps where minimizing risk matters most.

Watch out for: No programmatic Prompt creation, 30-day response TTL, continued API churn risk (this is OpenAI’s 4th major API paradigm shift).

Effort: Low–Medium · Cost: Same · Lock-in: High

🔀 Path 2: Switch to Another AI API Provider

Claude (best reasoning, long context), Gemini (free tier, multimodal), DeepSeek (1M context, thinking mode by default), Mistral (European hosting). Each has native tool use/function calling. Requires API integration changes but not architectural rewrites.

Choose this when: You want to reduce single-vendor dependency, need specific capabilities (long context, multimodal, EU hosting), or were already considering alternatives after repeated OpenAI API changes.

Watch out for: Different API shapes require code changes. Some features (code interpreter, web search) may not have direct equivalents. Evaluate each provider’s tool use implementation carefully.

Effort: Medium · Cost: Varies (often cheaper) · Lock-in: Medium

🌐 Path 3: Go Provider-Agnostic

Use an agent framework (LangChain, LlamaIndex, CrewAI) or multi-provider abstraction (OpenRouter, LiteLLM) to decouple from any single API. Switch models without code changes. Run open-source models locally for full control.

Choose this when: You want to future-proof against another deprecation cycle, need to compare providers on cost/quality, or want the ability to run models locally for privacy or cost reasons.

Watch out for: Abstraction layers add complexity and latency. Framework-specific lock-in replaces API-specific lock-in. Self-hosted models require ML ops capability.

Effort: Medium–High · Cost: Lowest at scale · Lock-in: Low

5. Alternatives

Direct API Alternatives

5 AI API providers compared. The tier and paid-rate columns are read from each provider's record in our index at request time, and the stability ratings come from our stability dashboard. A dash means the record carries no paid rate; the link goes to the vendor's own pricing page.

Provider Recorded Tier Tool Use Paid Rate Stability
OpenAI (Responses API) Pay-as-you-go Native function calling + MCP gpt-6-luna $0.10/$0.50 – gpt-6-astra $10.00/$50.00 per MTok volatile
  • OpenAI does not use data sent to the API to train its models unless you opt in to share data. (From developers.openai.com/api/docs/guides/your-data, read 2026-10-06.)
  • Abuse monitoring logs may contain prompts and responses and are kept for up to 30 days by default. (From developers.openai.com/api/docs/guides/your-data, read 2026-10-06.)
  • New API accounts use prepaid billing. Purchased credits expire after 1 year and are non-refundable unless required by law, contract, or an approved exception. (From help.openai.com/en/articles/8264644-how-can-i-set-up-prepaid-billing, read 2026-10-06.)
  • Without ChatGPT subscriber credits, OpenAI's API needs a payment method and a credit purchase of at least $5 before you use it. (From help.openai.com/en/articles/8264644-setting-up-and-managing-prepaid-api-billing, read 2026-10-22.)
  • OpenAI's API credits for ChatGPT subscribers need an active, paid ChatGPT Pro or Business subscription at least seven days old. The owner of the subscription or Business workspace links it to one API organization they own, using the same email. ChatGPT Plus, Go, Free and Enterprise plans, free trials and complimentary subscriptions are not eligible. No API payment method is needed while the credits last. Unused credits expire at the end of the billing cycle and do not roll over. They do not cover Codex used with an API key. (From developers.openai.com/api/docs/guides/chatgpt-api-credits, read 2026-10-29.)
  • gpt-6.2-astra, gpt-6.2-sol, gpt-6.1-terra and gpt-6-luna have 2-hour and 8-hour background completion windows. The 2-hour window is 30% below Standard. The 8-hour window is 40% below Standard. These windows are in public beta. They require Standard processing. They share the Batch/Background Rate Limit with Batch API jobs. (From developers.openai.com/api/docs/pricing, read 2026-11-18.)
Anthropic Claude Pay-as-you-go Native tool use Claude Haiku 5.5 $0.10/$0.50 – Claude Fable 5.1 $10/$50 per MTok volatile
  • Monthly credits with Claude Max, Ultra and Team plans pay for the Claude API, Claude Managed Agents and the Claude Agent SDK. That includes `claude -p` run with an API key from the Console organization linked to the plan. `claude -p` run while signed in with the plan uses the plan's usage limits, not the credits. The credits do not pay for interactive Claude Code, runs started by the Claude Code GitHub Action, an IDE extension or the desktop app, extra usage, or third-party apps billed through linked billing. They expire at the end of each billing cycle. New subscribers can claim them after seven days on the plan, with no payment method. Free, Pro and Enterprise plans are not eligible. (From support.claude.com/en/articles/17154008-monthly-api-credits-for-max-ultra-and-team-plans, read 2026-11-18.)
  • Anthropic's Claude API needs prepaid credits, bought with a payment method, before you use it. (From support.claude.com/en/articles/8977456-how-do-i-pay-for-my-claude-api-usage, read 2026-10-22.)
Google Gemini Free (Reduced) Native function calling Gemini 3.1 Flash-Lite $0.25/$1.50 – Gemini 3.1 Pro Preview $2/$12 per MTok watch
  • 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.)
DeepSeek Pay-as-you-go Native function calling DeepSeek-V4.1-Flash $0.30/$1.20, DeepSeek-V4-Pro $1.32/$3.96 per MTok volatile
  • Since November 1, 2026, 00:00 Beijing time (16:00 UTC on October 31), server-side Search and Fetch calls via DeepSeek Harness, Claude Code, or Codex are billed at $0.75 per 1,000 successful requests; they were free until then. (From api-docs.deepseek.com/quick_start/pricing, read 2026-11-01.)
  • DeepSeek's Search API and Fetch API are standalone APIs billed per request. Search finds web pages for a query; Fetch retrieves one web page as Markdown. Each costs $0.75 per 1,000 successful requests. Failed requests, empty search results, and timeouts are not billed. (From api-docs.deepseek.com/guides/search, read 2026-10-17.)
  • DeepSeek's Terms of Use, which also govern its API, let DeepSeek use inputs and outputs, de-identified and "to a minimal extent", to develop or improve its services and the technologies behind them. The terms say a user opts out by turning off "Improve the model for everyone". (From cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html, read 2026-10-23.)
  • DeepSeek's privacy policy says it collects, processes and stores personal data in the People's Republic of China. (From cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html, read 2026-10-23.)
Mistral AI Pay-as-you-go Native function calling Ministral 3 $0.1/$0.1 – Mistral Medium 3.5 $1.5/$7.5 per MTok volatile
  • Since August 2026, Mistral offers Mistral AI Studio and its API only to business customers. Its consumer terms for personal use exclude both, and API use falls under the Commercial Terms for an organization, company or other entity. (From legal.mistral.ai/terms/row-consumer-terms, read 2026-10-11.)
  • A business can use Mistral's API in a commercial product; individuals cannot use the API for personal projects. (From legal.mistral.ai/terms/commercial-terms-of-service, read 2026-10-11.)
  • Models whose names start with "labs-" are free of charge once you have API access; Mistral calls them experimental and not for production. (From docs.mistral.ai/inference/model-lifecycle, read 2026-11-12.)
  • In Free mode, Mistral may use your Studio inputs and outputs to train its models, and you can opt out at any time. (From help.mistral.ai/en/articles/347617-do-you-use-my-user-data-to-train-your-artificial-intelligence-models, read 2026-11-12.)
  • Mistral may train on inputs and outputs from Labs or Preview Models, which its Commercial Terms define as experimental, preview or pre-release models identified as such, including those with the "labs" prefix, even if you opted out or use zero data retention. Mistral Large 4 is a public preview. (From legal.mistral.ai/terms/commercial-terms-of-service, read 2026-10-11.)
  • Pay-as-you-go uses prepaid credits. Your monthly usage is consumed first. API and Vibe Code requests beyond it fail with HTTP 402 if your credit balance is zero. Add credits to restore access. Auto Recharge charges your payment method when your balance falls below a set threshold. (From docs.mistral.ai/admin/billing-usage/billing, read 2026-11-04.)

Agent Frameworks

Provider-agnostic frameworks that abstract away the underlying LLM API. Use these to avoid single-vendor lock-in and switch models without code changes.

Framework Type License Languages State Management
LangChain / LangGraph Agent framework MIT Python, JS/TS Built-in (LangGraph checkpointer)
LlamaIndex RAG + agents MIT Python, TS Workflow-based
CrewAI Multi-agent MIT Python Task-based crew state
AutoGen (Microsoft) Multi-agent CC-BY-4.0 Python, .NET Conversation-based
Vercel AI SDK Streaming toolkit Apache 2.0 TypeScript React state / server actions

Wire-Compatible Bridges

Drop-in replacements that mimic the Assistants API endpoint structure, letting you migrate with minimal code changes.

Bridge Approach Status Effort
Ragwalla Drop-in Assistants API replacement — same endpoints, backed by Responses API Active, maintained Minimal
DataStax astra-assistants-api Assistants API wire-compatible server backed by Astra DB + any LLM Active, open-source Low
Wire-compatible bridges explained: These services implement the same HTTP endpoints and request/response shapes as the Assistants API, so your existing client code works with just a base URL change. Ragwalla proxies to OpenAI’s Responses API under the hood, preserving the familiar Assistants interface. DataStax astra-assistants-api is open-source and backs the API with Astra DB, letting you swap in any LLM provider. Both are useful as interim solutions while you plan a full migration.

6. Token Cost Comparison

Every figure below is read from the provider's record in our index at request time — the model name included. Each provider contributes the cheapest and the dearest model its record carries a rate for. The last column prices 100M input plus 100M output tokens in a month at that rate. Per-tool charges are separate and are covered under the table.

Provider Model Input (per MTok) Output (per MTok) 100M in + 100M out
OpenAI (Responses API) gpt-6-luna $0.10 $0.50 $60
OpenAI (Responses API) gpt-6-astra $10.00 $50.00 $6,000
Anthropic Claude Claude Haiku 5.5 $0.10 $0.50 $60
Anthropic Claude Claude Fable 5.1 $10 $50 $6,000
Google Gemini Gemini 3.1 Flash-Lite $0.25 $1.50 $175
Google Gemini Gemini 3.1 Pro Preview $2 $12 $1,400
DeepSeek DeepSeek-V4.1-Flash $0.30 $1.20 $150
DeepSeek DeepSeek-V4-Pro $1.32 $3.96 $528
Mistral AI Ministral 3 $0.1 $0.1 $20
Mistral AI Mistral Medium 3.5 $1.5 $7.5 $900
The hidden cost of Responses API tools: token rates are not the whole bill. From OpenAI’s pricing documentation, read on 2026-09-27: file search storage is $0.10 per GB per day with the first GB free, and the file search tool call is billed separately at $2.50 per 1,000 calls. Hosted Shell and Code Interpreter run on containers charged per 20-minute session, from $0.03 at 1 GB to $1.92 at 64 GB, with a five-minute minimum. Web search is $10.00 per 1,000 calls, with retrieved content billed as input tokens at the model’s own rate, or $25.00 per 1,000 calls for the preview tool on non-reasoning models, whose search content tokens are free. We hold no record for any of these, so nothing re-verifies them — that date is when we read them. Claude processes PDFs with no per-file charge and Gemini includes grounding in the base token price.
Scale economics: at 100M input and 100M output tokens a month, the cheapest rate our index holds for these providers is Ministral 3 at $20, and the dearest is Claude Fable 5.1 at $6,000 — 300× the bill for the same traffic. Capability differs with it; the spread is the reason to measure your own workload rather than pick on price alone.

7. OpenAI Pricing Change Timeline

OpenAI’s stability rating is volatile. We’ve tracked 19 changes. Pattern: repeated free tier erosion and API paradigm shifts.

Date Change Impact
effective Oct 23, 2026 The OpenAI API keeps running. The gpt-3.5-turbo, gpt-4, gpt-4-turbo, gpt-4-1106-preview, gpt-4.1-nano, gpt-4o-2024-05-13, gpt-image-1, o1, o1-pro, o3-mini and o4-mini models, with their fine-tuned versions, can be called until October 23, 2026, when OpenAI shuts down their access. OpenAI names a substitute for each model: gpt-3.5-turbo to gpt-5.6-terra; gpt-4, gpt-4-turbo, gpt-4-1106-preview and gpt-4o-2024-05-13 to gpt-5.6-sol; gpt-4.1-nano to gpt-5.6-luna; gpt-image-1 to gpt-image-2.5-sunburst or gpt-image-2.5-flare; o1 to gpt-5.6-sol; o1-pro to gpt-5.6-sol with reasoning.mode: pro; o3-mini to gpt-5.6-sol; o4-mini to gpt-5.6-terra. Source ↗ MEDIUM
effective Sep 28, 2026 gpt-3.5-turbo-instruct, babbage-002, davinci-002 and gpt-3.5-turbo-1106 shut down in the OpenAI API on September 28, 2026. OpenAI names gpt-5.6-terra as the replacement. Source ↗ LOW
effective Sep 24, 2026 Sora 2 video generation in the OpenAI API shut down on September 24, 2026: OpenAI removed the Videos API and the sora-2 and sora-2-pro models, and names no replacement. Source ↗ MEDIUM
effective Aug 26, 2026 Assistants API deprecated, full shutdown August 26, 2026. Developers must migrate to Responses API + Conversations API Source ↗ HIGH
effective May 12, 2026 dall-e-2 and dall-e-3 were removed from the OpenAI API on 2026-05-12. OpenAI names gpt-image-2, gpt-image-1 or gpt-image-1-mini as substitutes, and gpt-image-1 itself shuts down on 2026-10-23. Source ↗ HIGH
effective May 12, 2026 The Realtime API beta was removed on 2026-05-12; OpenAI's generally available Realtime API replaces it. Source ↗ HIGH
effective Mar 20, 2024 Between 2024-03-13 and 2024-03-20, OpenAI stopped giving new API accounts the $5 free trial credit, which could be used during an account's first 3 months. New accounts now have to buy prepaid credits, $5 minimum, to use any paid model. Source ↗ HIGH

Upcoming Changes

These are announced changes that have not taken effect yet, listed soonest first.

Date Change Impact
effective Nov 30, 2026 The OpenAI API keeps running. Agent Builder, the Evals platform and the v1/prompts API can be used until November 30, 2026. Existing evals become read-only on October 31, 2026. ChatKit remains available. Migrate from Agent Builder to the Agents SDK or ChatGPT Workspace Agents. Migrate from Evals to Promptfoo. Move reusable prompt content into application code. Source ↗ LOW
effective Dec 1, 2026 The OpenAI API keeps running. The gpt-image-1-mini, gpt-image-1.5 and chatgpt-image-latest models can be called until December 1, 2026. OpenAI's substitutes are gpt-image-2.6-sunburst or gpt-image-2.6-flare. Source ↗ LOW
effective Dec 1, 2026 OpenAI's promotional pricing for gpt-5.6-sol on the OpenAI API ends on November 30, 2026. Until then, gpt-5.6-sol costs $4.00 per million input tokens, $0.40 per million cached input tokens and $20.00 per million output tokens on Standard processing, and $2.00, $0.20 and $10.00 on Flex and Batch. OpenAI has not said what gpt-5.6-sol costs from December 1, 2026. OpenAI's note covers gpt-5.6-sol on the OpenAI API only and says nothing about Azure OpenAI. Source ↗ MEDIUM
effective Dec 11, 2026 The OpenAI API keeps running. gpt-5 and gpt-5-2025-08-07 are available until December 11, 2026, replaced by gpt-6.1-sol. gpt-5-mini and gpt-5-mini-2025-08-07 are available until December 11, 2026, replaced by gpt-6.1-terra. gpt-5-nano and gpt-5-nano-2025-08-07 are available until December 11, 2026, replaced by gpt-6-luna. gpt-5-pro and gpt-5-pro-2025-10-06 are available until December 11, 2026, replaced by gpt-6.1-sol with reasoning.mode: pro in the Responses API. o3 and o3-2025-04-16 are available until December 11, 2026, replaced by gpt-6.1-sol. o3-pro and o3-pro-2025-06-10 are available until December 11, 2026, replaced by gpt-6.1-sol with reasoning.mode: pro in the Responses API. o3-mini is a different model from o3, and OpenAI shut it down on October 23. Source ↗ MEDIUM
effective Jan 6, 2027 The OpenAI API keeps running. The tts-1, tts-1-hd, gpt-4o-mini-tts-2025-03-20 and gpt-4o-mini-tts-2025-12-15 models can be called until January 6, 2027. OpenAI's replacement for each: gpt-realtime-2.1-mini. Source ↗ LOW
effective Jan 6, 2027 The OpenAI API keeps running. From January 6, 2027, no organization can create new fine-tuning jobs on OpenAI's self-serve fine-tuning platform. Since May 7, 2026, organizations that never ran fine-tuning cannot create jobs. Since July 2, 2026, organizations that have not run inference on a fine-tuned model in the past 60 days cannot either. Fine-tuned models keep serving inference until their base model is deprecated. Source ↗ MEDIUM
effective Jan 14, 2027 The OpenAI API keeps running. The gpt-4o (gpt-4o-2024-08-06 and gpt-4o-2024-11-20), gpt-4o-mini, gpt-4.1 and gpt-4.1-mini models, with their fine-tuned versions, can be called until January 14, 2027, when OpenAI shuts down their access. OpenAI names gpt-6-luna as the replacement for all four. Source ↗ MEDIUM
effective Jan 20, 2027 The OpenAI API keeps running. The gpt-realtime, gpt-4o-realtime, gpt-realtime-mini, gpt-4o-mini-realtime, gpt-audio, gpt-4o-audio, gpt-audio-mini, gpt-4o-mini-audio and gpt-4o-mini-transcribe-2025-03-20 models can be called until January 20, 2027. OpenAI's substitutes: gpt-realtime-2.1 for gpt-realtime and gpt-4o-realtime; gpt-realtime-2.1-mini for gpt-realtime-mini and gpt-4o-mini-realtime; gpt-audio-1.5 for gpt-audio, gpt-4o-audio, gpt-audio-mini and gpt-4o-mini-audio; gpt-4o-mini-transcribe-2025-12-15 for gpt-4o-mini-transcribe-2025-03-20. Source ↗ LOW
effective Feb 1, 2027 The OpenAI API keeps running. The gpt-5.5-cyber-preview model can be called until February 1, 2027, when OpenAI removes it from the API. OpenAI tells users to move to the most capable cyber model available to them. Source ↗ LOW
effective Feb 10, 2027 The OpenAI API keeps running. The gpt-image-2 model and its gpt-image-2-2026-04-21 snapshot can be called until February 10, 2027, when OpenAI removes them from the API. OpenAI names gpt-image-2.6-sunburst or gpt-image-2.6-flare as the replacement. Source ↗ MEDIUM
effective Feb 26, 2027 The OpenAI API keeps running. The whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe and gpt-4o-transcribe-diarize models can be called until February 26, 2027. OpenAI's substitute for each is gpt-live-transcribe or gpt-transcribe. The /v1/audio/translations endpoint also shuts down on February 26, 2027. OpenAI's replacement is /v1/realtime/translations. Source ↗ MEDIUM
effective Apr 1, 2027 The OpenAI API keeps running. The gpt-5.3-codex, gpt-5.1 and gpt-5.4-nano models can be called until April 1, 2027. OpenAI's replacements: gpt-6-sol for gpt-5.3-codex and gpt-5.1; gpt-6-luna for gpt-5.4-nano. Source ↗ MEDIUM

Recommendations by Use Case

Which Path for Which Developer

Least code changes:

OpenAI Responses API — direct migration with feature parity. Start here unless you have a reason to leave.

Zero code changes (interim):

Ragwalla or DataStax astra-assistants-api — wire-compatible bridges that keep your Assistants API code working while you plan a real migration.

Best reasoning & long context:

Anthropic Claude — Claude Fable 5.1 $10/$50 per MTok at the top of the lineup our record holds. Best for complex multi-step agent workflows.

Free tier available:

Google Gemini — free tier with rate limits, multimodal input. Best for prototyping and low-volume production.

Cheapest at scale:

Mistral AI — Ministral 3 $0.1/$0.1 per MTok, the lowest paid rate our index carries for any provider on this page.

Future-proof against deprecations:

LangChain/LangGraph or OpenRouter — abstract the LLM layer so you can switch providers without code changes.

Enterprise with compliance needs:

Mistral AI (EU hosting) or AutoGen (Azure/Microsoft ecosystem). Both offer function calling with data residency options.

Methodology

How we track this data: AgentDeals monitors free tier changes across 1,498 developer tools in 60 categories. Stability ratings are computed from our deal changes database — OpenAI is classified as volatile based on 19 tracked changes including free tier removal, limit reductions, and API deprecation.

Migration complexity ratings are based on the scope of code changes required: Low (API shape change only), Medium (some logic restructuring), High (architectural changes to state management or data flow).

Cost data is from official vendor pricing pages. Compiled 2026-04-03, not re-checked since. Tool-specific costs (file search, web search) are Responses API additions not present in the Assistants API.

For real-time data, use our stability dashboard, Atom feed, or MCP server. Full dataset available via REST API.

Related Guides