Benchmarks / Coding
FrontierSWE v2
Cognition's ultra-long-horizon software engineering benchmark: 34 tasks graded 0-100 on implementation, performance and research; mean@5 with a 20-hour budget per task in the Proximus harness at max reasoning effort.
FrontierSWE v2
Higher is better · best effort per model
full table below
| # | Model | Effort | Score |
|---|---|---|---|
| 1 | GPT-6 Astra | max | 65.5% |
| 2 | Opus 5.5 | max | 62.3% |
| 3 | Fable 5.1 | max | 56.3% |
| 4 | Gemini 4 | max | 55.0% |
| 5 | Opus 5 | max | 52.0% |
| 6 | Fable 5 | max | 47.0% |
| 7 | Haiku 5.5 | max | 43.8% |
| 8 | GPT-5.6 Sol | max | 32.2% |
| 9 | GLM 5.3 | max | 30.2% |
| 10 | Grok 4.7 | max | 29.5% |
| 11 | Kimi K3 | max | 25.9% |
| 12 | Muse Spark 1.3 | max | 25.8% |
| 13 | Grok 4.6 | max | 25.3% |
| 14 | Gemini 3.7 Flash | max | 20.3% |
| 15 | Gemini 3.8 Flash | max | 19.6% |
| 16 | GLM 5.3 Flash | max | 18.1% |
| 17 | Qwen 3.8 Max | max | 17.8% |
| 18 | DeepSeek V4 Flash | max | 14.8% |
| 19 | Muse Spark 1.2 | max | 12.0% |
| 20 | Inkling | max | 4.1% |