OpenAI shuts down GPT-3.5, GPT-4 and o1 in the API tomorrow

OpenAI’s deprecations page lists a hard shutdown of 23 October for a cluster of its oldest API models: gpt-3.5-turbo, gpt-4 (the gpt-4-0613 snapshot), gpt-4-turbo, gpt-4-1106-preview, o1, o1-pro, o3-mini and o4-mini. Fine-tuned models built on gpt-3.5-turbo, gpt-4, o4-mini, babbage-002 or davinci-002 are retired on the same date, since a fine-tune can’t outlive its base model.

OpenAI’s recommended replacements split by size: gpt-5.6-sol for the GPT-4 and o1 family, gpt-5.6-terra for gpt-3.5-turbo and o4-mini. Everything here has had at least six months’ notice under OpenAI’s standard policy, but gpt-3.5-turbo and the original gpt-4 are also the two model names most likely to be hardcoded in old scripts, tutorials and forgotten cron jobs that nobody has touched since the API first shipped.

gpt-3.5-turbo -> gpt-5.6-terra
gpt-4         -> gpt-5.6-sol
o1, o3-mini   -> gpt-5.6-sol
o4-mini       -> gpt-5.6-terra

Why it matters: any request to these model IDs starts failing on 23 October, not degrading gracefully. If you’ve got a pinned model string anywhere in production, including inside a fine-tuned model, today is the day to check, not tomorrow.

The caveat: gpt-image-1 appears on the same deprecations page but its shutdown is 1 December, not 23 October, so don’t lump it in with this batch.