SWE Atlas - Test Writing
Code · 2026-05-07
Score on SWE Atlas test-writing tasks that write production-grade tests for repository behavior.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 55.6 |
| Opus 4.8 | 49.6 |
| GPT-5.6 Sol | 45.9 |
| GPT-5.4 | 44.4 |
| GPT-5.5 | 42.6 |
| GLM-5.2 | 41.5 |
| Muse Spark 1.1 | 41.5 |
| GPT-5.3-Codex | 39.0 |
| Opus 4.7 | 38.5 |
| Opus 4.6 | 36.7 |
| Sonnet 4.6 | 31.8 |
| Muse Spark | 31.1 |
| Gemini 3 Flash Preview | 30.3 |
| Gemini 3.1 Pro Preview | 29.8 |
| GLM-5 | 28.7 |
| DeepSeek-V4-Pro | 27.1 |
| Kimi K2.5 | 25.8 |
| MiniMax M2.5 | 18.6 |