SWE-Bench ProMax (OpenHands): leaderboard
SWE-Bench ProMax (2026) gives coding agents 170 large refactoring tasks from real commits in Python, Java, TypeScript, Go, C, C++ and Rust, averaging 11.4 modified files, with rewritten issue descriptions and manually reviewed tests. Resolve rate is the percentage of instances whose patch passes every test; this board runs each model in the OpenHands scaffold with a 300-step and 10 US dollar limit. Higher is better.
Metric: Resolve rate (%, Pass@1; 170 repository-level refactoring instances from real commits (average 11.4 modified files), resolved when the patch passes every test of the manually reviewed suite; OpenHands scaffold, step limit 300, cost limit 10 US dollars per instance). Source: arxiv.org. Saturation forecast: Around March 2028. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 41.2 | #176 |
| 2 | Claude Sonnet 4.6 | 38.8 | #161 |
| 3 | GLM-5 | 36.5 | #260 |
| 4 | Kimi K2.5 | 32.9 | #256 |
| 5 | Gemini 3 Pro | 19.4 | #137 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=swe-bench-promax-openhands · How It Works · Data refreshed daily, snapshot 2026-11-13.