Read from the providers' own deprecation pages on 8 November 2026. Dates change, so check the pages before you act on this.

Date Model ID What the provider page says
30 Nov 2026 claude-sonnet-4-5-20250929 Anthropic: replacement claude-sonnet-5-5. "Requests to models past the retirement date will fail."
1 Dec 2026 gemini-2.5-flash Google: calls route automatically to gemini-3.5-flash-lite, billed at that model's rates
1 Dec 2026 gemini-2.5-pro Google: calls route automatically to gemini-3.8-flash, billed at that model's rates
11 Dec 2026 gpt-5, o3 OpenAI: replacement gpt-5.6-sol
11 Dec 2026 gpt-5-mini / gpt-5-nano OpenAI: replacements gpt-5.6-terra / gpt-5.6-luna

Sources: Anthropic, Google, OpenAI.

The detail that is easy to miss

Anthropic says calls to a retired model fail. Google's 1 December change is different: the old model name keeps answering, but Google routes the request to another model and bills it at that model's rates. Google also tells you to read the breaking API changes in the new model's migration guide before 1 December. Nothing errors, so nothing alerts you. If you rely on gemini-2.5-pro for quality, the call is answered by a Flash model after that date. Google also says its shutdown dates in the table are the earliest possible dates, and that gemini-2.5-flash-lite is not deprecated.

Find the IDs in your code

Plain grep -w is not enough. It matches gemini-2.5-flash-lite (not deprecated), gpt-5.6-sol (the replacement) and o3-mini. This pattern skips them; I tested it on a sample file with those lines (GNU grep, LC_ALL=C.UTF-8):

LC_ALL=C.UTF-8 grep -rnP '(?<![\w.-])(claude-sonnet-4-5-20250929|gemini-2\.5-flash|gemini-2\.5-pro|gpt-5|gpt-5-mini|gpt-5-nano|o3)(?![\w.-])' --exclude-dir=node_modules --exclude-dir=.git .

Check config files, Docker and CI settings, .env.example, eval suites and prompt tests too, not only code. It will not find IDs built at run time.

OpenAI also says you can call GET /v1/models and read each model's optional shutdown_date field.

Before the date

  1. Run the grep on every repo and config folder.
  2. Re-run your structured-output and tool-call tests on the replacement. The replacement is a different model, so its output can differ.
  3. Compare cost per request on the replacement. Google bills the rerouted calls at the new model's rates.
  4. Pin the replacement ID yourself rather than waiting for a reroute.

Disclosure

I sell a paid offline Python scanner that does the matching, prints file:line, the shutdown date and the named replacement, and includes these dates and a migration checklist: LLM Model Retirement Scanner on Gumroad, $19. You do not need it; the three pages and the grep above are free. I am not affiliated with OpenAI, Anthropic or Google.