Calibration
SciCode → Vals LiveCodeBench
Vals LiveCodeBench is estimated from SciCode with a linear curve fitted on 7 models measured on both: y = 0.2490·x + 0.7387, R² = 0.75, cross-validated error 0.8 pp. It is used for 21 estimates.
| Estimated model | SciCode | Vals LiveCodeBench | Source |
|---|---|---|---|
| A.X K2 | 41.0% | 84.1% | estimated ± 0.8 pp, low confidence |
| Agents-A1-4B | 29.6% | 81.2% | estimated ± 0.8 pp, low confidence |
| Granite 4.2 30B | 38.8% | 83.5% | estimated ± 0.8 pp, low confidence |
| Granite 4.2 3B | 24.1% | 79.9% | estimated ± 0.8 pp, low confidence |
| Granite 4.2 8B | 36.1% | 82.9% | estimated ± 0.8 pp, low confidence |
| Hy3 Preview | 41.2% | 84.1% | estimated ± 0.8 pp, low confidence |
| K-EXAONE 2.0 | 37.4% | 83.2% | estimated ± 0.8 pp, low confidence |
| Kimi K2.5 | 48.7% | 86.0% | estimated ± 0.8 pp, medium confidence |
| Ling 2.6 Flash | 27.0% | 80.6% | estimated ± 0.8 pp, low confidence |
| Ling 3.0 Flash FP8 | 40.4% | 83.9% | estimated ± 0.8 pp, low confidence |
| Mercury 2.5 | 38.0% | 83.3% | estimated ± 0.8 pp, low confidence |
| MiniCPM5-2B | 26.3% | 80.4% | estimated ± 0.8 pp, low confidence |
| Muse Glimmer 30B | 43.6% | 84.7% | estimated ± 0.8 pp, medium confidence |
| Nemotron 3.5 Lightning 30B A3B NVFP4 | 31.4% | 81.7% | estimated ± 0.8 pp, low confidence |
| Nemotron 3 Nano Omni 30B A3B | 32.0% | 81.8% | estimated ± 0.8 pp, low confidence |
| Qwen 3.6 Max (preview) | 47.0% | 85.6% | estimated ± 0.8 pp, medium confidence |
| Qwen3.7 Plus | 51.3% | 86.6% | estimated ± 0.8 pp, medium confidence |
| Beam | 49.7% | 86.2% | estimated ± 0.8 pp, medium confidence |
| Sakana Fugu | 60.1% | 88.8% | estimated ± 0.8 pp, low confidence |
| Sakana Fugu-Ultra | 58.7% | 88.5% | estimated ± 0.8 pp, low confidence |
| Step 5 Preview | 58.9% | 88.5% | estimated ± 0.8 pp, low confidence |