FrontierSWE (Dominance)
Code · 2026-04-16
FrontierSWE is an ultra-long-horizon software-engineering benchmark scored by model performance on difficult repository-level tasks.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5 | 90.0 |
| Kimi K3 | 81.2 |
| Opus 4.8 | 78.0 |
| Grok 4.5 | 78.0 |
| GPT-5.5 | 76.0 |
| GLM-5.2 | 72.0 |
| GPT-5.6 Sol | 71.3 |
| Opus 4.7 | 66.0 |
| Opus 4.6 | 59.0 |
| GPT-5.4 | 56.0 |
| Gemini 3.1 Pro Preview | 41.0 |
| Composer 2.5 | 38.0 |
| GLM-5.1 | 32.0 |
| DeepSeek-V4-Pro | 31.0 |
| Kimi K2.6 | 28.0 |
| Kimi K2.5 | 27.0 |
| Qwen3.6 Plus (2026-04-02) | 24.0 |