Model Details
o1-mini
Quick answer: o1-mini is OpenAI's September 2024 cost-efficient reasoning model, a smaller sibling to o1-preview optimized for STEM tasks like coding and math. On Codeforces-style competitive-programming evals it reaches up to 1,820 Elo (93.4th percentile) under the DeepSeek-R1 comparison protocol, and 1,578 Elo (89.2th percentile) under the independent CodeElo protocol.
At a Glance
Where o1-mini leads
- Strong Codeforces-style competitive-programming performance for its cost tier
- Lower cost and latency than o1-preview, targeted at STEM reasoning tasks
Where it lags
- General knowledge/world-knowledge tasks lag behind o1-preview and full o1
- Proprietary: no open weights
Benchmark Leaderboards
This model isn’t on any benchmark leaderboard yet.