SWE-bench Live: leaderboard
Live SWE-bench-style benchmark with recent real GitHub software issues, reducing contamination while measuring repository-level repair ability.
Metric: Resolved (%). Source: swe-bench-live.github.io. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | TianxiCode + Deepseek-v4.1-Flash | 71 | |
| 2 | Slingshot-v3.4.0 + GPT-5.6-Sol | 70.33 | |
| 3 | AMI Agent + Claude-4.6-Opus | 63 | |
| 4 | agav-agent + GPT-5.5 (Medium) | 62 | |
| 5 | Slingshot-v3.3.0 + GPT-5.6-Sol | 44.67 | |
| 6 | SWE-agent + Claude-4.5-Sonnet | 40 | |
| 7 | AiWork.Code + Claude-4.8-Opus (xhigh) | 34 | |
| 8 | SWE-agent + GPT5-Thinking (Medium) | 30.4 | |
| 9 | Slingshot-v3.1.0 + Claude-4.6-Opus | 25 | |
| 10 | OpenHands + Qwen3-Coder-480B-A35B | 24.67 |
Interactive version: theaggregate.ai/benchmark?slug=swe-bench-live · How It Works · Data refreshed daily, snapshot 2026-10-20.