Calibration
BrowseComp → AutomationBench
AutomationBench is estimated from BrowseComp with a inverse Michaelis–Menten curve fitted on 7 models measured on both: y = 0.24873·(x − 0.0000) / (0.0000 + 1.4849 − x), R² = 0.31, cross-validated error 10.3 pp. It is used for 14 estimates.
| Estimated model | BrowseComp | AutomationBench | Source |
|---|---|---|---|
| Agents-A1 | 75.5% | 25.7% | estimated ± 10.3 pp, low confidence |
| Agents-A1-4B | 66.8% | 20.3% | estimated ± 10.3 pp, low confidence |
| dots3-note Preview | 83.3% | 31.8% | estimated ± 10.3 pp, low confidence |
| GPT-5.2 | 65.8% | 19.8% | estimated ± 10.3 pp, low confidence |
| GPT-5.4 Pro | 89.3% | 37.5% | estimated ± 10.3 pp, low confidence |
| GPT-5.5 Pro | 90.1% | 38.4% | estimated ± 10.3 pp, low confidence |
| LongCat-Flash-Lite-Sparse | 48.6% | 12.1% | estimated ± 10.3 pp, low confidence |
| Ornith-1.5-35B-A3B | 67.6% | 20.8% | estimated ± 10.3 pp, low confidence |
| Ornith-1.5-397B | 86.6% | 34.8% | estimated ± 10.3 pp, low confidence |
| Ornith-1.5-9B | 56.4% | 15.2% | estimated ± 10.3 pp, low confidence |
| Qwen3.5-27B | 61.0% | 17.3% | estimated ± 10.3 pp, low confidence |
| Qwen3.5-35B-A3B | 61.0% | 17.3% | estimated ± 10.3 pp, low confidence |
| Qwen3.5 397B | 62.0% | 17.8% | estimated ± 10.3 pp, low confidence |
| Beam | 77.4% | 27.1% | estimated ± 10.3 pp, low confidence |