o3-mini
2025-01-31 · Publisher : OpenAI · Domain : Artificial Intelligence
OpenAI launches o3-mini, its most cost-efficient compact reasoning model, optimized for science, mathematics and code, available in ChatGPT and the API the same day.
What changes
- First small OpenAI reasoning model with function calling, structured outputs and developer messages.
- Three reasoning effort levels (low, medium, high); o3-mini-high is offered in ChatGPT to paying subscribers.
- Replaces o1-mini in ChatGPT, with a limit raised from 50 to 150 messages per day for Plus and Team.
- First time a reasoning model is available to free users (“Reason” option).
- Built-in web search as a prototype.
### Specifications - Context 200,000 tokens, maximum output 100,000 tokens, text-only input, text output. - Knowledge cutoff: October 1, 2023. - API identifier: o3-mini (snapshot o3-mini-2025-01-31), through Chat Completions, Responses and Batch. ### Pricing Pricing per million tokens (USD), as listed in the API documentation. | | Input | Cached input | Output | | --- | --- | --- | --- | | o3-mini | 1.10 | 0.55 | 4.40 | ### Benchmarks Scores published by the vendor in the announcement text (the other results appear in charts). | Measure | Result | | --- | --- | | FrontierMath, high effort, with Python | more than 32% on the first attempt | | SWE-bench Verified, high, Agentless scaffold | 39% | | SWE-bench Verified, high, internal scaffold | 61% | | Expert preference versus o1-mini | 56% | | Major errors versus o1-mini | 39% fewer | ### Key takeaways o3-mini makes reasoning accessible at low cost, with responses 24% faster than o1-mini (7.7 s versus 10.16 s on average). It suits high-volume technical uses.
Links
Tags : LLM, OpenAI, o-series, Raisonnement, Mini