FrontierSWE V2 Worst@5
Code
FrontierSWE V2 mean of the worst normalized reward per task across up to five graded trials. This is not a confidence-interval bound.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 55.1 |
| Claude Opus 5.5 | 52.0 |
| Gemini 4 Argon | 44.5 |
| Claude Opus 5 | 39.1 |
| Claude Haiku 5.5 | 34.5 |
| GPT-5.6 Sol | 22.2 |
| GLM-5.3 | 17.7 |
| Grok 4.7 | 15.9 |
| Kimi K3 | 13.9 |
| Grok 4.6 | 12.9 |
| Muse Spark 1.3 | 12.6 |
| Gemini 3.7 Flash | 10.3 |
| Gemini 3.8 Flash | 9.5 |
| GLM-5.3 Flash | 8.6 |
| Qwen3.8 Max (0902) | 8.1 |
| Muse Spark 1.2 | 6.8 |
| Inkling | 0.1 |