OpenAI: o1-preview

OpenAI o1-preview was a research preview of OpenAI's first o-series reasoning model. OpenAI's model page describes it as trained with reinforcement learning to handle complex reasoning, producing an internal chain of thought before it answers. The page now labels it shut down. According to OpenAI's deprecations page, developers using o1-preview and o1-mini were notified on April 28, 2025 that both would be removed from the API, and o1-preview shut down on July 28, 2025 with o3 as the recommended replacement.

Status at a glance

The sibling o1-mini followed with a shutdown date of October 27, 2025 and o4-mini as the recommended replacement. See o1-mini in this catalog.

Specs While It Was Available

The model page lists these specs for the retired model:

These prices are historical, since the model no longer accepts requests.

Key Features at the Time

• Reasoning before answering: the model spent tokens on an internal chain of thought to work through complex problems before replying.
• Preview status: OpenAI's model page calls it a research preview.
• Higher cost than general chat models: the listed output price of $60 per 1M tokens reflected the extra reasoning work.

Use Cases Today

• Finding out why a legacy o1-preview call fails.
• Studying the first generation of OpenAI reasoning models and how pricing and limits changed afterward.
• Choosing a migration target for a prompt set written for o1-preview.

A limitation: since the model is shut down, none of its behaviour can be tested today, and OpenAI's pages give no benchmark numbers for the retired model.

Migration Steps

1. Search your code for o1-preview and o1-preview-2024-09-12.
2. Replace it with o3, as OpenAI recommends, or with a current model from OpenAI's models list.
3. Re-run your evals, because reasoning models differ in output length, cost and behaviour.

FAQ

Is o1-preview still available?

No. OpenAI's deprecations page lists its shutdown as July 28, 2025.

What replaced o1-preview?

OpenAI's recommended replacement is o3.

What was the context window of o1-preview?

128,000 tokens, with a maximum output of 32,768 tokens.

How much did o1-preview cost?

OpenAI's model page lists $15 per 1M input tokens, $7.50 cached input and $60 per 1M output tokens.

Alternatives

• OpenAI o1-mini: the smaller o1 model, also retired.
• GPT-5.6: a newer OpenAI frontier model in this catalog.
• GPT-5.2: an earlier GPT-5 series model.
• DeepSeek R1: an open-weight reasoning model.

Conclusion

o1-preview was the preview that introduced OpenAI's o-series reasoning approach, and it has been shut down since July 28, 2025. Migrate to o3 or a newer model, and keep OpenAI's deprecations page handy for future changes. More models are in the models category.