Calibration
SWE-bench Pro → CursorBench 3.1
CursorBench 3.1 is estimated from SWE-bench Pro with a Michaelis–Menten + offset curve fitted on 5 models measured on both: y = 0.0000 + 2.0000·x / (1.59619 + x), R² = 0.73, cross-validated error 5.6 pp. It is used for 35 estimates.
| Estimated model | SWE-bench Pro | CursorBench 3.1 | Source |
|---|---|---|---|
| Claude Mythos 5 | 80.3% | 66.9% | estimated ± 5.6 pp, low confidence |
| Sakana Fugu-Ultra | 73.7% | 63.2% | estimated ± 5.6 pp, low confidence |
| Qwen3.8 Max | 67.7% | 59.6% | estimated ± 5.6 pp, low confidence |
| Beam 1.0 501B-A23B FP8 | 65.9% | 58.4% | estimated ± 5.6 pp, low confidence |
| Hy4 preview | 65.7% | 58.3% | estimated ± 5.6 pp, low confidence |
| Beam 1.0 501B-A23B NVFP4 | 65.6% | 58.3% | estimated ± 5.6 pp, low confidence |
| Ornith-1.5-397B | 65.1% | 57.9% | estimated ± 5.6 pp, low confidence |
| Qwen3.8-Omni-Flash | 63.3% | 56.8% | estimated ± 5.6 pp, low confidence |
| Ornith-1.0-397B | 62.2% | 56.1% | estimated ± 5.6 pp, low confidence |
| Sage-A1.5 | 61.2% | 55.4% | estimated ± 5.6 pp, low confidence |
| dots3-note Preview | 61.0% | 55.3% | estimated ± 5.6 pp, low confidence |
| Atria Dawn Preview | 59.6% | 54.4% | estimated ± 5.6 pp, low confidence |
| Ornith-1.5-35B-A3B | 59.6% | 54.4% | estimated ± 5.6 pp, low confidence |
| Laguna S 2.1 | 59.4% | 54.2% | estimated ± 5.6 pp, low confidence |
| Sakana Fugu | 59.0% | 54.0% | estimated ± 5.6 pp, low confidence |
| Qwen 3.6 Max (preview) | 57.3% | 52.8% | estimated ± 5.6 pp, low confidence |
| Claude Opus 4.5 | 57.1% | 52.7% | estimated ± 5.6 pp, low confidence |
| GPT-5.3 Codex | 56.8% | 52.5% | estimated ± 5.6 pp, low confidence |
| MiMo-V2.5 | 56.1% | 52.0% | estimated ± 5.6 pp, low confidence |
| GPT-5.2 | 55.6% | 51.7% | estimated ± 5.6 pp, low confidence |
| GLM-5 | 55.1% | 51.3% | estimated ± 5.6 pp, low confidence |
| Spark-X2.6-4B | 54.3% | 50.8% | estimated ± 5.6 pp, low confidence |
| Claude Opus 4.6 | 53.4% | 50.1% | estimated ± 5.6 pp, low confidence |
| MAI-Thinking-1 | 52.8% | 49.7% | estimated ± 5.6 pp, low confidence |
| Grok 4.20 | 51.8% | 49.0% | estimated ± 5.6 pp, low confidence |
| Kimi K2.5 | 50.7% | 48.2% | estimated ± 5.6 pp, low confidence |
| Ornith-1.0-35B | 50.4% | 48.0% | estimated ± 5.6 pp, low confidence |
| Qwen3.6-35B-A3B | 49.5% | 47.3% | estimated ± 5.6 pp, low confidence |
| Laguna M.1 | 49.2% | 47.1% | estimated ± 5.6 pp, low confidence |
| Laguna XS 2.1 | 47.6% | 45.9% | estimated ± 5.6 pp, low confidence |
| Ornith-1.5-9B | 47.5% | 45.9% | estimated ± 5.6 pp, low confidence |
| Laguna XS.2 | 46.3% | 45.0% | estimated ± 5.6 pp, low confidence |
| Ornith-1.0-9B | 42.9% | 42.4% | estimated ± 5.6 pp, low confidence |
| LongCat-Flash-Lite-Sparse | 40.6% | 40.6% | estimated ± 5.6 pp, low confidence |
| Granite 4.2 30B | 33.3% | 34.5% | estimated ± 5.6 pp, low confidence |
Fitted on
| Model | SWE-bench Pro | CursorBench 3.1 |
|---|---|---|
| Claude Fable 5 | 80.0% | 70.6% |
| Claude Opus 4.8 | 69.2% | 58.4% |
| Gemini 3.5 Flash | 55.1% | 49.8% |
| GPT-5.5 | 58.6% | 59.2% |
| Kimi K2.6 | 58.6% | 47.6% |