SWE-bench Verified - OpenAI 477-Task Subset
Code · 2025-04-16
The 477-task SWE-bench Verified subset used for OpenAI's o3 and o4-mini release evaluation. It excludes 23 of the 500 parent tasks that could not be run on OpenAI infrastructure and must be distinguished from full-set results.
Top models (higher is better)
| Model | Score |
|---|---|
| o3 | 69.1 |
| O4 Mini | 68.1 |