Calibration
SciCode → SWE-bench Verified
SWE-bench Verified is estimated from SciCode with a Hill curve fitted on 19 models measured on both: y = 0.3879 + (0.8864 − 0.3879)·x^6.00 / (0.38521^6.00 + x^6.00), R² = 0.90, cross-validated error 5.1 pp. It is used for 7 estimates.
| Estimated model | SciCode | SWE-bench Verified | Source |
|---|---|---|---|
| A.X K2 | 41.0% | 68.3% | estimated ± 5.1 pp, medium confidence |
| Ling 3.0 Flash FP8 | 40.4% | 67.2% | estimated ± 5.1 pp, medium confidence |
| Mercury 2.5 | 38.0% | 62.7% | estimated ± 5.1 pp, medium confidence |
| Nemotron 3 Nano Omni 30B A3B | 32.0% | 51.1% | estimated ± 5.1 pp, medium confidence |
| Agents-A1-4B | 29.6% | 47.3% | estimated ± 5.1 pp, medium confidence |
| Ling 2.6 Flash | 27.0% | 44.1% | estimated ± 5.1 pp, medium confidence |
| Granite 4.2 3B | 24.1% | 41.6% | estimated ± 5.1 pp, low confidence |
Fitted on
| Model | SciCode | SWE-bench Verified |
|---|---|---|
| Granite 4.2 30B | 38.8% | 57.0% |
| Granite 4.2 8B | 36.1% | 47.7% |
| Hy3 Preview | 41.2% | 74.4% |
| Inkling-Small | 48.7% | 80.2% |
| K-EXAONE 2.0 | 37.4% | 68.2% |
| Kimi K2.6 | 52.2% | 80.2% |
| Kimi K2.5 | 48.7% | 76.8% |
| LFM2.5-24B-A2B | 32.7% | 51.0% |
| LFM3-1.7B | 28.5% | 43.4% |
| MiniCPM5-2B | 26.3% | 46.4% |
| Muse Glimmer 30B | 43.6% | 76.0% |
| Nemotron 3.5 Lightning 30B A3B NVFP4 | 31.4% | 52.8% |
| Nemotron 3 Ultra | 44.6% | 71.9% |
| Qwen3.7 Max | 53.5% | 80.4% |
| Qwen3.7 Plus | 51.3% | 77.7% |
| Beam 1.0 501B-A23B | 49.2% | 81.6% |
| Beam 1.0 501B-A23B FP8 | 48.3% | 80.2% |
| Beam 1.0 501B-A23B NVFP4 | 46.9% | 80.7% |
| Spark-X2.6-4B | 36.8% | 62.8% |