OpenAI Sets January 14, 2027 Deadline for Five GPT-4 API Snapshots

Five GPT-4-series API snapshots and their fine-tuned versions are scheduled to retire on January 14, 2027. Illustration: MacDigest.

OpenAI will remove five GPT-4-series snapshots from its API on January 14, 2027, according to a deprecation notice published October 14. Fine-tuned versions of the affected models share the deadline. The notice gives no cutoff hour or time zone.

Check model IDs and aliases

The affected snapshots are:

• gpt-4o-2024-08-06
• gpt-4o-2024-11-20
• gpt-4o-mini-2024-07-18
• gpt-4.1-2025-04-14
• gpt-4.1-mini-2025-04-14

Search application code, environment settings, scheduled jobs and deployment configuration for those IDs, plus fine-tuned model IDs derived from them. Include scripts and integrations that run without someone opening a new session.

For each affected deployment, record the base model, any fine-tuned ID, the endpoint used and who owns the change. Check vendor-managed integrations separately: a model setting in your own code may not control the requests they make.

Check unversioned aliases too. OpenAI’s model pages currently map gpt-4o, gpt-4o-mini, gpt-4.1 and gpt-4.1-mini to snapshots on this list. The notice does not promise automatic rerouting to a replacement.

Fine-tuned deployments need extra work

OpenAI recommends gpt-6-luna for every model in the notice, but Luna’s model page lists fine-tuning as unsupported. Teams using custom fine-tunes need to evaluate how to reproduce the required behavior with a supported model.

Test the prompts, supplied context and examples your application needs. Compare results against the evaluations used for the existing fine-tune, including cases where a particular response format or tool choice matters. A model-string change alone does not establish that the replacement meets those requirements.

Test before switching production

There is a specific tool-calling constraint to check: GPT-6 Luna supports function calling through Chat Completions only when reasoning_effort is set to none. For reasoning with tools, OpenAI’s migration quickstart directs developers to the Responses API.

For reasoning effort other than none, OpenAI also says to remove temperature, top_p and top_logprobs. That same condition applies to logprobs in Chat Completions and message.output_text.logprobs in the Responses include list.

Run representative requests before changing production traffic. Check output formats, function arguments, error handling, latency and cost, then verify that scheduled work uses the replacement. Leave time to correct regressions before January 14.

This is separate from the GPT-5.5 retirement through ChatGPT sign-in covered earlier, whose notice explicitly excluded the OpenAI API.