OpenAI gives gpt-5.1 six months' notice, after 20 days for gpt-5.4-cyber

OpenAI gives gpt-5.1 six months' notice, after 20 days for gpt-5.4-cyber

OpenAI has put a shutdown date on three of its API models, with the full notice period its own policy promises. gpt-5.1, gpt-5.3-codex and gpt-5.4-nano "will be removed from the API on April 1, 2027, with six months' notice", OpenAI says in a deprecation notice dated October 1, 2026, three weeks after it gave gpt-5.4-cyber about 20 days.

OpenAI title card reading GPT-6 Sol and Luna, with a sun and a crescent moon

OpenAI points gpt-5.1 and gpt-5.3-codex at gpt-6-sol, and gpt-5.4-nano at gpt-6-luna. Image: OpenAI

The recommended replacements are not one-for-one. Both gpt-5.1 and the Codex model gpt-5.3-codex are pointed at gpt-6-sol, while the small gpt-5.4-nano is pointed at gpt-6-luna.

Six months is the floor OpenAI sets for generally available models. Specialised variants get at least three, and OpenAI's own worked example of a specialised variant is "Codex variants such as gpt-5.3-codex", so the Codex model has been given twice the notice it was owed.

Below both of those sits a third rung. Preview models, "identified by preview in the model name, may be retired with much shorter notice, such as 2 weeks", and OpenAI's advice is not to put business-critical work on one "unless you can migrate on short notice". None of the three models dated October 1 is a preview model, and OpenAI says impacted customers are always notified by email as well as in its documentation.

One model left on 20 days

Three weeks before this, the same policy produced a very different number. On September 11, OpenAI deprecated gpt-5.4-cyber for removal on October 1, 2026, which we covered when a cyber model was pulled on about 20 days' notice where the published policy sets months. OpenAI has not said why the two cases were handled differently, and the deprecation notice offers no reason for either figure.

Anthropic, for comparison, gave Claude Sonnet 4.5 61 days, one day above its own floor. On that scale, April 1, 2027 is generous.

One thing worth keeping straight: these are API model IDs, not the entries in ChatGPT's model picker. If you call OpenAI's models from your own code, a service or a plugin, these three are the ones with 2027 dates on them. If you only use ChatGPT, nothing in this notice changes what you see.

The queue behind these three is long

April 2027 is some way down the list. The v1/prompts API and reusable prompt objects shut down on November 30, 2026, and so do two products: Agent Builder, and the Evals platform, whose existing evals become read-only on October 31.

December 11, 2026 takes the older GPT-5 and o3 snapshots, gpt-5-2025-08-07 and gpt-5-pro-2025-10-06 among them, with GPT-5.6 models named as their replacements. Then the audio work goes. The legacy audio and realtime families shut down on January 20, 2027, and tts-1 and the other text-to-speech models announced on the same October 1 date go on January 6, 2027. The transcription models, whisper-1 among them, follow on February 26, 2027.

OpenAI leaves one door open after a shutdown, and only one. "In some cases, developers may be able to provision dedicated capacity for continued access after a model's shutdown date," it says, pointing anyone interested at its sales team. That is a sentence about possibility, not an offer, and OpenAI attaches no price or eligibility to it.

The practical reading is that anything still calling gpt-5.1 or gpt-5.3-codex has until April 1, 2027 to move, with a model family two generations newer waiting on the other side. Whether the next specialised variant gets six months or 20 days is the part OpenAI has not committed to.

Leave a Comment

Your email address will not be published. Required fields are marked *