Deadline guide · updated 13 Nov 2026
OpenAI model shutdowns: test the replacement before you switch
On this page
The short answer
- 11 Dec 2026: gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3 and o3-pro shut down. 14 Jan 2027: GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini. Source
- OpenAI names a replacement for each, and says to run representative requests against both models and compare outputs, latency and cost before switching.
- A replacement is a different model. Check format, policy, refusals and language, not just average quality. The free 8-case smoke test covers the usual breakages.
Key numbers
- OpenAI shuts down gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3 and o3-pro on 11 Dec 2026 (replacements: gpt-6.1-sol, gpt-6.1-terra, gpt-6-luna). (OpenAI deprecations page, 13 Nov 2026)
- Anthropic retires claude-sonnet-4-5-20250929 on 30 Nov 2026; the recommended replacement is claude-sonnet-5-5. (Anthropic model deprecations, 13 Nov 2026)
- From 1 Dec 2026, Gemini API requests to gemini-2.5-pro route to gemini-3.8-flash and are billed at its rates. (Gemini API deprecations, 13 Nov 2026)
Shutdown dates and replacements
| Shutdown | Notice | Model | OpenAI's recommended replacement |
|---|---|---|---|
| 11 Dec 2026 | 11 Jun 2026 | gpt-5 | gpt-6.1-sol |
| 11 Dec 2026 | 11 Jun 2026 | gpt-5-mini | gpt-6.1-terra |
| 11 Dec 2026 | 11 Jun 2026 | gpt-5-nano | gpt-6-luna |
| 11 Dec 2026 | 11 Jun 2026 | gpt-5-pro | gpt-6.1-sol with reasoning.mode: pro (Responses API) |
| 11 Dec 2026 | 11 Jun 2026 | o3 | gpt-6.1-sol |
| 11 Dec 2026 | 11 Jun 2026 | o3-pro | gpt-6.1-sol with reasoning.mode: pro (Responses API) |
| 14 Jan 2027 | 14 Oct 2026 | gpt-4o (2024-08-06 and 2024-11-20 snapshots) | gpt-6-luna |
| 14 Jan 2027 | 14 Oct 2026 | gpt-4o-mini | gpt-6-luna |
| 14 Jan 2027 | 14 Oct 2026 | gpt-4.1 | gpt-6-luna |
| 14 Jan 2027 | 14 Oct 2026 | gpt-4.1-mini | gpt-6-luna |
| 1 Apr 2027 | 1 Oct 2026 | gpt-5.1 | gpt-6-sol |
| 1 Apr 2027 | 1 Oct 2026 | gpt-5.3-codex | gpt-6-sol |
| 1 Apr 2027 | 1 Oct 2026 | gpt-5.4-nano | gpt-6-luna |
Text models only, from OpenAI's deprecations page as read on 13 Nov 2026. OpenAI says generally available models get at least 6 months' notice. The linked page is authoritative if dates change.
Why test before you switch
Your prompt was tuned, often without anyone noticing, to one model's habits: how long it writes, how strictly it follows format rules, when it refuses, how it handles a language switch or an injected instruction. A newer model can be better on average and still break one of those. That's a prompt regression caused by the model instead of the prompt.
A 4-step model migration test
- Collect 20–50 real cases. Use representative requests from your logs (remove personal data), plus the edge cases you know about: format requests, policy questions, refusals that shouldn't happen, other languages, prompt injection. Start from the smoke-test template or the support test cases.
- Write the expected behavior. For each case, describe what a good answer must do. If the current model's answers are good, use them as the baseline (“should match this answer in substance”). See how to write expected behavior.
- Run both models side by side. In Litmiq, select the current model (if the app offers it) and its replacement; the app offers gpt-6-astra, gpt-6.1-sol, gpt-6.1-terra, gpt-6-luna and gpt-5.4-mini. An LLM judge scores every output with a reason.
- Read every disagreement. Look at the cases where the two models' scores differ most, read both outputs, and fix the prompt where the new model needs clearer instructions. Re-run until the replacement is at least as good on the cases that matter.
What to compare
| Check | What to look for | How |
|---|---|---|
| Quality | Same or better on your cases; no new failures on hard rules | Judge scores, plus reading the outputs that differ |
| Format | Bullet limits, JSON-only replies, length | Cases that ask for a format, with strict expected behavior |
| Policy and safety | No invented policies or promises; injection still resisted | Policy and injection test cases |
| Refusals | Doesn't refuse normal requests the old model answered | Safe but unusual questions |
| Cost | Price per 1M input and output tokens, and output length | LLM API pricing 2026 calculator |
| Latency | Time to a complete answer for your typical request | Time a sample of real requests in your own stack |
Can a cheaper model do the same job?
A shutdown is a good moment to test one tier down. Run the same suite on the recommended replacement and on a cheaper model (for example gpt-6.1-terra and gpt-6-luna). If the cheaper model passes every hard rule and its scores are close, the saving may be worth it. Check prices on the LLM API pricing page and see how models did on our reliability benchmark. Don't decide from averages alone: one failure on a policy case can matter more than a point of average score.
Comparing more than two models? See compare OpenAI models for prices and a cost calculator.
Limits
- Litmiq runs OpenAI and Anthropic models only, single-turn, and doesn't measure latency. Measure latency in your own stack.
- A judge score is an estimate. Read the outputs, and check the judge against your own labels for decisions that matter (how).
- A model that's already shut down can't be run, so capture baseline outputs from the current model before its date.
FAQ
When is gpt-5 shut down?
OpenAI's deprecations page lists gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-pro, o3 and o3-pro for shutdown on 11 December 2026, with notice given on 11 June 2026. Read on 13 Nov 2026.
What replaces gpt-5-mini?
OpenAI's deprecations page lists gpt-6.1-terra as the recommended replacement for gpt-5-mini, gpt-6.1-sol for gpt-5 and o3, and gpt-6-luna for gpt-5-nano.
How much notice does OpenAI give before a shutdown?
OpenAI's deprecations page says generally available models get at least 6 months' notice before shutdown.
How do I know the replacement model behaves the same?
You don't until you test it. OpenAI's own guidance is to run representative requests against both models and compare outputs, latency and cost before switching production traffic. Run the same test cases on both, read the outputs that differ, and check format, policy and refusal behavior, not only average scores.
Can I test a model migration in Litmiq for free?
Litmiq is free and runs in your browser, but it uses your own OpenAI API key, so OpenAI bills you for the calls: one completion per case per model plus one judge call per output. A 20-case test on two models is 40 completions plus 40 judge calls.
Sources (read 13 Nov 2026)
- OpenAI API deprecations: shutdown dates, notice dates and recommended replacements above; at least 6 months' notice for generally available models; “Run representative requests against both models. Compare outputs, latency, and cost before switching production traffic.”
Also moving off OpenAI Evals? See how to export and migrate your OpenAI Evals before 30 Nov 2026.
Get model deprecation alerts
We email you when OpenAI, Anthropic or Google publish a new shutdown date for a model on our tracker. Sent by hand from hello@litmiq.com, a few times a month at most. Unsubscribe by replying.
We store only your email, through our form provider. See the privacy policy.
Test the replacement model
Load the 8-case smoke test, pick your current model and its replacement, and compare. Free, in your browser.
- Free, no signup
- Runs in your browser
- Your OpenAI key, sent only to OpenAI