Calibration
CharXiv → CharXiv w/o tools
CharXiv w/o tools is estimated from CharXiv with a offset logistic curve fitted on 14 models measured on both: y = 0.7759 + (1.2000 − 0.7759) / (1 + exp(−25.30·(x − 0.9772))), R² = 0.70, cross-validated error 2.6 pp. It is used for 4 estimates.
| Estimated model | CharXiv | CharXiv w/o tools | Source |
|---|---|---|---|
| Muse Spark 1.1 | 88.4% | 81.3% | estimated ± 2.6 pp, high confidence |
| Sakana Fugu-Ultra | 86.6% | 80.0% | estimated ± 2.6 pp, high confidence |
| Sakana Fugu | 85.1% | 79.3% | estimated ± 2.6 pp, high confidence |
| Nemotron 3 Nano Omni 30B A3B | 76.3% | 77.8% | estimated ± 2.6 pp, medium confidence |
Fitted on
| Model | CharXiv | CharXiv w/o tools |
|---|---|---|
| Claude Mythos 5 | 93.5% | 88.9% |
| Claude Opus 4.7 | 91.0% | 82.1% |
| Claude Opus 4.8 | 89.9% | 80.5% |
| Claude Sonnet 5 | 88.3% | 77.0% |
| Gemini 3.7 Flash | 88.7% | 84.5% |
| Inkling | 82.0% | 78.1% |
| Inkling 2 Small | 87.7% | 85.6% |
| Inkling-Small | 81.3% | 77.4% |
| Kimi K3.1 Pro | 93.4% | 89.0% |
| Kimi K3 | 91.3% | 84.8% |
| Qwen3.8-27B | 90.2% | 83.7% |
| Qwen3.8-Flash-Next | 90.6% | 84.6% |
| Qwen3.8 Max | 93.5% | 88.4% |
| Qwen3.8-Omni-Flash | 91.4% | 83.5% |