Calibration
CharXiv → ERQA
ERQA is estimated from CharXiv with a Michaelis–Menten curve fitted on 12 models measured on both: y = 2.0000·x / (1.62569 + x), R² = 0.68, cross-validated error 3.7 pp. It is used for 18 estimates.
| Estimated model | CharXiv | ERQA | Source |
|---|---|---|---|
| Claude Mythos 5 | 93.5% | 73.0% | estimated ± 3.7 pp, high confidence |
| Claude Opus 4.7 (Adaptive) | 91.0% | 71.8% | estimated ± 3.7 pp, high confidence |
| Claude Opus 4.8 | 89.9% | 71.2% | estimated ± 3.7 pp, high confidence |
| Claude Sonnet 4.6 | 77.4% | 64.5% | estimated ± 3.7 pp, high confidence |
| Claude Sonnet 5 | 88.3% | 70.4% | estimated ± 3.7 pp, high confidence |
| Command A+ | 52.7% | 49.0% | estimated ± 3.7 pp, medium confidence |
| Gemini 3.1 Flash-Lite | 73.2% | 62.1% | estimated ± 3.7 pp, high confidence |
| Gemini 3.5 Flash | 84.2% | 68.2% | estimated ± 3.7 pp, high confidence |
| Gemini 3.7 Flash | 88.7% | 70.6% | estimated ± 3.7 pp, high confidence |
| GLM-5.3-Flash | 89.4% | 71.0% | estimated ± 3.7 pp, high confidence |
| Inkling | 82.0% | 67.1% | estimated ± 3.7 pp, high confidence |
| Inkling-Small | 81.3% | 66.7% | estimated ± 3.7 pp, high confidence |
| MiMo-V2.5 | 81.0% | 66.5% | estimated ± 3.7 pp, high confidence |
| Muse Glimmer 30B | 78.8% | 65.3% | estimated ± 3.7 pp, high confidence |
| Muse Spark 1.1 | 88.4% | 70.4% | estimated ± 3.7 pp, high confidence |
| Nemotron 3 Nano Omni 30B A3B | 76.3% | 63.9% | estimated ± 3.7 pp, high confidence |
| Sakana Fugu | 85.1% | 68.7% | estimated ± 3.7 pp, high confidence |
| Sakana Fugu-Ultra | 86.6% | 69.5% | estimated ± 3.7 pp, high confidence |