SWE Atlas - Codebase QnA
Code · 2026-03-04
Score on SWE Atlas codebase QnA tasks that require runtime analysis and multi-file reasoning.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 63.2 |
| Claude Fable 5.1 | 60.0 |
| GPT-6 Astra | 59.1 |
| Opus 4.8 | 57.3 |
| DeepSeek V4.1 Flash | 53.5 |
| Kimi K3 | 52.7 |
| Nanbeige4.3-33B | 50.8 |
| Claude Sonnet 5.5 | 50.3 |
| Grok 4.6 | 50.3 |
| Sonnet 5 | 49.5 |
| Claude Opus 5.5 | 48.9 |
| Claude Haiku 5.5 | 48.7 |
| GPT-6.1 Sol | 48.4 |
| GLM-5.2 | 48.1 |
| Gemini 3.8 Flash | 47.0 |
| Opus 4.7 | 47.0 |
| Grok 4.7 | 47.0 |
| Laguna S 2.1 | 46.2 |
| GPT-5.6 Luna | 46.0 |
| GPT-5.6 Sol | 46.0 |
| GPT-5.5 | 45.4 |
| GPT-5.6 Terra | 45.2 |
| GPT-6 Sol | 44.1 |
| MiMo-V2.6-Pro | 42.7 |
| GLM-5.3 | 42.5 |
| Muse Spark 1.1 | 42.2 |
| Grok 4.5 | 41.9 |
| GPT-5.4 | 40.8 |
| Claude Fable 5 | 39.0 |
| MiniMax M3 | 37.9 |
| Sonnet 4.6 | 35.5 |
| GPT-6 Luna | 33.9 |
| Opus 4.6 | 33.3 |
| Kimi K2.7 Code | 33.1 |
| GPT-5.3-Codex | 32.6 |
| Gemini 3.6 Flash | 32.0 |
| Beam 1.0 501B A23B | 31.8 |
| GPT-5.6 Sol Pro | 30.9 |
| DeepSeek-V4-Pro | 27.2 |
| Muse Spark | 24.2 |