LiveBench Python: leaderboard
The Python slice of LiveBench's agentic coding category, where the model works in a real repository over many turns to resolve an issue taken from a public GitHub project. Runs use the Mini SWE Agent scaffold with a 250 step limit and are graded by the Multi-SWE-Bench harness, which runs the project's own tests. The score is a percentage of issues resolved, higher is better.
Metric: Score. Source: livebench.ai. Saturation forecast: Around 2029. 71 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V4.1 Flash (Max) | 90 | #36 (DeepSeek V4.1 Flash) |
| 2 | Qwen 3.8 27B | 75 | #105 |
| 3 | Claude Opus 5 (Max) | 75 | #12 (Claude Opus 5) |
| 4 | DeepSeek-V4-Flash-Vision-Exp | 75 | #76 |
| 5 | Qwen3.8-Flash-Next | 70 | #72 |
| 6 | Claude Fable 5.1 (Max) | 70 | #9 (Claude Fable 5.1) |
| 7 | Claude Opus 5.5 (Max) | 70 | #3 (Claude Opus 5.5) |
| 8 | Claude Opus 5.5 (xHigh) | 70 | #3 (Claude Opus 5.5) |
| 9 | Kimi K3 | 65 | #30 |
| 10 | GPT-6 Astra (Max) | 65 | #6 (GPT-6 Astra) |
| 11 | Claude Fable 5 (Max) | 65 | #16 (Claude Fable 5) |
| 12 | GLM-5.3 Flash | 60 | #65 |
| 13 | Qwen 3.8 Max | 60 | #35 |
| 14 | GLM-5.3 | 60 | #39 |
| 15 | Grok 4.5 | 60 | #48 |
No result here: #8 Gemini 4 Argon, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=livebench-python · How It Works · Data refreshed daily, snapshot 2026-10-20.