Calibration
SWE Multilingual → MMLU-ProX
MMLU-ProX is estimated from SWE Multilingual with a Hill curve fitted on 8 models measured on both: y = 0.8012 + (1.2000 − 0.8012)·x^6.00 / (1.04769^6.00 + x^6.00), R² = 0.52, cross-validated error 1.3 pp. It is used for 35 estimates.
| Estimated model | SWE Multilingual | MMLU-ProX | Source |
|---|---|---|---|
| Ling 3.1 Flash | 88.5% | 90.7% | estimated ± 1.3 pp, medium confidence |
| Claude Opus 4.8 | 84.4% | 88.7% | estimated ± 1.3 pp, medium confidence |
| Hy4 preview | 82.9% | 88.0% | estimated ± 1.3 pp, medium confidence |
| Qwen3.8-Flash-Next | 81.0% | 87.1% | estimated ± 1.3 pp, medium confidence |
| Qwen3.8-Omni-Flash | 80.5% | 86.9% | estimated ± 1.3 pp, medium confidence |
| Composer 2.5 | 79.8% | 86.6% | estimated ± 1.3 pp, medium confidence |
| Ornith-1.5-397B | 79.6% | 86.6% | estimated ± 1.3 pp, medium confidence |
| Ornith-1.0-397B | 78.9% | 86.3% | estimated ± 1.3 pp, medium confidence |
| Laguna S 2.1 | 78.5% | 86.1% | estimated ± 1.3 pp, medium confidence |
| Claude Sonnet 5 | 78.3% | 86.0% | estimated ± 1.3 pp, high confidence |
| Beam 1.0 501B-A23B | 78.3% | 86.0% | estimated ± 1.3 pp, high confidence |
| Beam 1.0 501B-A23B FP8 | 78.1% | 86.0% | estimated ± 1.3 pp, high confidence |
| Grok 4.5 | 78.0% | 85.9% | estimated ± 1.3 pp, high confidence |
| SWE-1.7 | 77.8% | 85.8% | estimated ± 1.3 pp, high confidence |
| Beam 1.0 501B-A23B NVFP4 | 77.5% | 85.7% | estimated ± 1.3 pp, high confidence |
| Kimi K2.6 | 76.7% | 85.4% | estimated ± 1.3 pp, high confidence |
| MiniMax M2.7 | 76.5% | 85.4% | estimated ± 1.3 pp, high confidence |
| DeepSeek V4 Pro 0813 | 76.2% | 85.3% | estimated ± 1.3 pp, high confidence |
| dots3-note Preview | 75.7% | 85.1% | estimated ± 1.3 pp, high confidence |
| Composer 2 | 73.7% | 84.4% | estimated ± 1.3 pp, high confidence |
| DeepSeek V4 Flash 0731 | 73.3% | 84.3% | estimated ± 1.3 pp, high confidence |
| Sage-A1.5 | 72.7% | 84.1% | estimated ± 1.3 pp, high confidence |
| Ling 3.0 Flash | 72.4% | 84.0% | estimated ± 1.3 pp, high confidence |
| MiMo-V2-Flash | 71.7% | 83.8% | estimated ± 1.3 pp, high confidence |
| Ornith-1.5-35B-A3B | 71.4% | 83.8% | estimated ± 1.3 pp, high confidence |
| Qwen3.6-27B | 71.3% | 83.7% | estimated ± 1.3 pp, high confidence |
| Ornith-1.0-35B | 69.3% | 83.2% | estimated ± 1.3 pp, high confidence |
| Nemotron 3.5 Super VL 120B A12B | 67.7% | 82.8% | estimated ± 1.3 pp, high confidence |
| Qwen3.6-35B-A3B | 67.2% | 82.7% | estimated ± 1.3 pp, medium confidence |
| Nemotron 3.5 Super VL 120B A12B NVFP4 | 67.0% | 82.7% | estimated ± 1.3 pp, medium confidence |
| Spark-X2.6-4B | 64.7% | 82.2% | estimated ± 1.3 pp, medium confidence |
| Laguna M.1 | 63.1% | 81.9% | estimated ± 1.3 pp, medium confidence |
| Laguna XS 2.1 | 63.1% | 81.9% | estimated ± 1.3 pp, medium confidence |
| LongCat-Flash-Lite-Sparse | 59.3% | 81.4% | estimated ± 1.3 pp, medium confidence |
| Laguna XS.2 | 57.7% | 81.2% | estimated ± 1.3 pp, medium confidence |
Fitted on
| Model | SWE Multilingual | MMLU-ProX |
|---|---|---|
| Claude Opus 4.5 | 77.5% | 85.7% |
| GLM-5 | 73.3% | 83.1% |
| Kimi K2.5 | 73.0% | 82.3% |
| Nemotron 3 Ultra | 67.7% | 83.0% |
| Qwen3.5 397B | 69.3% | 84.7% |
| Qwen3.6 Plus | 73.8% | 84.7% |
| Qwen3.7 Max | 78.3% | 87.0% |
| Qwen3.7 Plus | 75.8% | 85.4% |