Calibration
ResearchClawBench → DeepSearchQA
DeepSearchQA is estimated from ResearchClawBench with a Michaelis–Menten curve fitted on 6 models measured on both: y = 1.4123·x / (0.12758 + x), R² = 0.42, cross-validated error 10.2 pp. It is used for 3 estimates.
| Estimated model | ResearchClawBench | DeepSearchQA | Source |
|---|---|---|---|
| MiMo-V2.5 | 16.9% | 80.5% | estimated ± 10.2 pp, low confidence |
| MiMo-V2-Pro | 15.3% | 77.0% | estimated ± 10.2 pp, low confidence |
| Grok 4.1 | 13.5% | 72.6% | estimated ± 10.2 pp, low confidence |
Fitted on
| Model | ResearchClawBench | DeepSearchQA |
|---|---|---|
| Claude Opus 4.6 | 19.9% | 73.7% |
| Claude Opus 4.8 | 21.1% | 93.1% |
| Gemini 3.1 Pro | 13.3% | 69.7% |
| GPT-5.4 | 15.3% | 73.6% |
| Kimi K2.6 | 18.0% | 92.5% |
| Kimi K2.5 | 14.0% | 77.1% |