Multi-SWE-bench Python (MagentLess): leaderboard
The Python split of the Multi-SWE-bench leaderboard (ByteDance Seed, 2025): the 500 human-validated issues of SWE-bench Verified, run under the benchmark's own agents. The agent edits the repository to resolve each issue, and a fix counts only when the project's tests pass. Scored as the percentage of issues resolved under MagentLess, the benchmark team's multilingual port of Agentless; higher is better.
Metric: Resolved rate (%). Source: multi-swe-bench.github.io. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro (Preview 03-25) | 49 | #234 |
| 2 | O1 (2024-12-17) | 48.2 | #310 |
| 3 | O3 Mini (2025-01-31) (High) | 46.4 | #486 (O3 Mini (2025-01-31)) |
| 4 | Claude 3.7 Sonnet (20250219) | 44.6 | #399 |
| 5 | Claude 3.5 Sonnet (20241022) | 42.4 | #635 |
| 6 | DeepSeek R1 | 42.2 | #467 |
| 7 | DeepSeek V3 | 41 | #705 |
| 8 | Llama 4 Maverick | 37.8 | #1105 |
| 9 | GPT-4o (2024-11-20) | 36.2 | #864 |
| 10 | Qwen 2.5 72B Instruct | 26.8 | #1093 |
| 11 | Doubao-1.5-Pro | 26.2 | #585 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=multi-swe-bench-python-magentless · How It Works · Data refreshed daily, snapshot 2026-11-12.