- Home
- Benchmarks
- SWE-Bench Pro
Agents · long tasks · reference
SWE-Bench Pro
Longer, harder software tasks across several languages, from Scale AI. Closer to a day of real engineering work than a single bug fix.
- Models
- 24
- By makers
- 24
- Latest
- Oct 13
% of tasks resolved. v marks a score the model’s maker reported. Prices are FullsendAI’s per million tokens, blended 3 input : 1 output.
About the data
From Scale AI SWE-Bench Pro. No explicit licence on the leaderboard data; Scale's general site terms apply. The benchmark code is MIT.