SWE Atlas: leaderboard
SWE Atlas Codebase QnA evaluates LLMs on deep code comprehension and question answering across real-world software repositories. (Source: benchmarklist.com, self-reported.)
Metric: Score (mixed sources). Source: benchmarklist.com. 62 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3.8 Max 0902 | 66.3 | |
| 2 | Claude Opus 5 (max) | 66 | #12 (Claude Opus 5) |
| 3 | DeepSeek V4 Pro 0813 | 66 | |
| 4 | Kimi K3 | 66 | #30 |
| 5 | Hy4 preview claude-code | 64 | |
| 6 | Step 5 Preview (high) | 63.6 | #44 (Step 5 Preview) |
| 7 | Opus 5 (Claude Code) xHigh (xHigh) | 63.17 | |
| 8 | Claude Opus 5 | 62 | #12 |
| 9 | Qwen3.8 Max | 62 | |
| 10 | Qwen3.8-2.4T-A95B (xhigh effort, Claude Code) (xHigh) | 62 |
Interactive version: theaggregate.ai/benchmark?slug=swe-atlas · How It Works · Data refreshed daily, snapshot 2026-10-20.