OpenAI is removing a large batch of older chat and reasoning models from the API on 23 October 2026 — including gpt-4 (gpt-4-0613), gpt-3.5-turbo, gpt-4-turbo, gpt-4.1-nano, the gpt-4o-2024-05-13 snapshot, o1, o1-pro, o3-mini and o4-mini — directing developers to the gpt-5.6 family

OpenAI’s API deprecations page sets a single shutdown date of 23 October 2026 for a large batch of older chat and reasoning models. After that date, a request that names one of these model ids returns an error instead of a completion.

The models being removed on 23 October:

• gpt-3.5-turbo (gpt-3.5-turbo-0125)
• gpt-4 (gpt-4-0613), gpt-4-1106-preview
• gpt-4-turbo (gpt-4-turbo-2024-04-09)
• gpt-4.1-nano (gpt-4.1-nano-2025-04-14)
• gpt-4o-2024-05-13 (the dated snapshot of gpt-4o)
• gpt-image-1
• o1 (o1-2024-12-17), o1-pro
• o3-mini, o4-mini
• fine-tuned models built on gpt-3.5-turbo, gpt-4, gpt-4.1-nano, babbage-002 and davinci-002

OpenAI points developers to its gpt-5.6 family — gpt-5.6-sol, gpt-5.6-terra and gpt-5.6-luna — with the exact per-model mapping listed on the deprecations page.

Read the version numbers carefully

The list is specific, and it is easy to over-read. The general gpt-4o, gpt-4.1 and gpt-4.5 models are not being removed — only the ids above. The retiring gpt-4.1-nano is the nano tier, not gpt-4.1 itself. And it is the dated gpt-4o-2024-05-13 snapshot that goes, not gpt-4o. If you pinned a general model name you are probably fine; if you pinned an old dated snapshot or one of the retiring families, you are not.

Where this hurts

For anyone who selects a model dynamically or just uses the current default, this is a non-event. It bites where a model id is hardcoded and forgotten — a wrapper class that defaults to gpt-4, an environment variable set two years ago, a batch job pinned to gpt-3.5-turbo, a prompt-eval config that still names o1. Those references stop resolving on 23 October, and the failure surfaces wherever that code runs, not where you would think to look.

If you maintain fine-tuned models, note that a fine-tune retires with its base model: fine-tunes on gpt-3.5-turbo, gpt-4, gpt-4.1-nano, babbage-002 or davinci-002 need to be re-tuned on a supported base before the date.

What to do

Before 23 October 2026, grep for the literal model ids across source, environment files, and any prompt or evaluation configs — not just your main application code — and repoint each to its gpt-5.6 replacement. Because swapping a model can change output shape, latency and cost, re-run your evaluations against the replacement rather than treating this as a blind find-and-replace. The lastcall CLI fingerprints the retiring ids (gpt-3.5-turbo, gpt-4-0613, gpt-4-turbo, gpt-4.1-nano, gpt-4o-2024-05-13, o1-pro, o3-mini, o4-mini and the dated gpt-4-1106-preview), so a pinned model that would break in October surfaces as a warning in your terminal now — never matching the surviving gpt-4o, gpt-4.1 or gpt-4.5.

This is one wave of a longer schedule. OpenAI’s page also lists the GPT-5 dated snapshots, o3 and o3-pro for removal on 11 December 2026, the older image models on 1 December, and the legacy audio, realtime and transcription models into 2027 — alongside the already-scheduled Sora 2 video models on 24 September. Anything that names an OpenAI model by string is worth a periodic check against the deprecations page.