Calibration
Gert Labs → ResearchClawBench
ResearchClawBench is estimated from Gert Labs with a linear curve fitted on 16 models measured on both: y = 0.1581·x + 0.0796, R² = 0.41, cross-validated error 2.2 pp. It is used for 13 estimates.
| Estimated model | Gert Labs | ResearchClawBench | Source |
|---|---|---|---|
| Claude 4 Sonnet | 39.7% | 14.2% | estimated ± 2.2 pp, medium confidence |
| Claude Sonnet 4.5 | 48.5% | 15.6% | estimated ± 2.2 pp, medium confidence |
| Gemini 3.1 Flash-Lite | 38.5% | 14.0% | estimated ± 2.2 pp, medium confidence |
| Gemini 3 Pro | 63.2% | 18.0% | estimated ± 2.2 pp, medium confidence |
| GPT-4.1 | 25.7% | 12.0% | estimated ± 2.2 pp, low confidence |
| GPT-5.1-Codex | 49.7% | 15.8% | estimated ± 2.2 pp, medium confidence |
| GPT-5.2-Codex | 51.8% | 16.1% | estimated ± 2.2 pp, medium confidence |
| GPT-5.3 Codex | 57.5% | 17.0% | estimated ± 2.2 pp, medium confidence |
| Grok 4 | 42.3% | 14.7% | estimated ± 2.2 pp, medium confidence |
| Grok 4.1 Fast | 47.3% | 15.4% | estimated ± 2.2 pp, medium confidence |
| Grok 4.20 | 38.4% | 14.0% | estimated ± 2.2 pp, medium confidence |
| Grok Build 0.1 | 49.2% | 15.7% | estimated ± 2.2 pp, medium confidence |
| Qwen3 Max | 43.7% | 14.9% | estimated ± 2.2 pp, medium confidence |