o3 and o4-mini
o3 and o4-mini are trained to use tools agentically and to reason over images, setting records on Codeforces, SWE-bench and MMMU.
First OpenAI reasoning models that decide when to call web search, Python and image tools inside their chain of thought, and think with images. o4-mini hits 99.5% on AIME 2025 with a Python interpreter. API returns reasoning summaries; o3 costs $10/$40 per million tokens (cut 80% on 2025-06-10).
- Date
- Wednesday, 16 April 2025
- Lab
- OpenAI
- Kind
- model
- Access
- closed API
- Price
- o3 $10/$40; o4-mini $1.10/$4.40 per 1M tokens (2025-04-16)
Figures
| Measure | Value | Measured by |
|---|---|---|
| AIME 2025, o4-mini with Python interpreter | 99.5% company calls the benchmark saturated | company |
| o3 API price (input / output per 1M tokens) | $10 / $40 200K context, 100K max output; cut to $2 / $8 on 2025-06-10 | company |
| o4-mini API price (input / output per 1M tokens) | $1.10 / $4.40 same as o3-mini | company |
See openai-o3-announce for the December 2024 preview. o3 and o4-mini snapshots are scheduled for API removal on 2026-12-11 (deprecations page, announced 2026-06-11).
Sources
- openai.com/index/introducing-o3-and-o4-mini/
- simonwillison.net/2025/Apr/16/introducing-openai-o3-and-o4-mini/
- developers.openai.com/api/docs/changelog
This record was checked against its sources on 6 October 2026. How we check