Calibration
BrowseComp → AA Agentic Index
AA Agentic Index is estimated from BrowseComp with a inverse Michaelis–Menten curve fitted on 21 models measured on both: y = 0.19446·(x − 0.0000) / (0.0000 + 1.2581 − x), R² = 0.72, cross-validated error 8.2 pp. It is used for 10 estimates.
| Estimated model | BrowseComp | AA Agentic Index | Source |
|---|---|---|---|
| LongCat-2.5 | 91.9% | 52.7% | estimated ± 8.2 pp, medium confidence |
| GPT-5.5 Pro | 90.1% | 49.1% | estimated ± 8.2 pp, medium confidence |
| GPT-5.4 Pro | 89.3% | 47.6% | estimated ± 8.2 pp, medium confidence |
| Sage-A1.5 | 81.7% | 36.0% | estimated ± 8.2 pp, medium confidence |
| Ling 3.0 Flash Sante | 80.2% | 34.2% | estimated ± 8.2 pp, medium confidence |
| Agents-A1 | 75.5% | 29.2% | estimated ± 8.2 pp, medium confidence |
| Agents-A1-4B | 66.8% | 22.0% | estimated ± 8.2 pp, medium confidence |
| Sarvam 105B | 49.5% | 12.6% | estimated ± 8.2 pp, medium confidence |
| LongCat-Flash-Lite-Sparse | 48.6% | 12.2% | estimated ± 8.2 pp, medium confidence |
| Sarvam 30B | 35.5% | 7.6% | estimated ± 8.2 pp, low confidence |
Fitted on
| Model | BrowseComp | AA Agentic Index |
|---|---|---|
| Claude Opus 4.7 | 79.3% | 39.5% |
| Claude Opus 4.8 | 84.3% | 42.6% |
| Claude Opus 5 | 90.8% | 56.2% |
| Claude Sonnet 5 | 84.7% | 44.3% |
| DeepSeek V4 Flash 0731 | 73.2% | 41.7% |
| DeepSeek V4 Pro | 83.4% | 49.6% |
| GLM-5.1 | 68.0% | 25.2% |
| GPT-5.5 | 84.4% | 37.3% |
| GPT-5.6 Luna | 83.3% | 42.7% |
| GPT-5.6 Sol | 92.2% | 50.5% |
| GPT-5.6 Terra | 87.5% | 43.7% |
| GPT-6 Astra | 91.5% | 51.5% |
| Inkling | 77.1% | 24.3% |
| Inkling-Small | 77.4% | 25.0% |
| Kimi K2.6 | 83.2% | 22.1% |
| Kimi K3 | 91.2% | 50.6% |
| Ling 3.0 Flash | 72.2% | 21.0% |
| MiniMax M3 | 83.5% | 30.8% |
| Nemotron 3.5 Lightning 30B A3B NVFP4 | 36.8% | 6.1% |
| Nemotron 3 Ultra | 44.4% | 21.7% |
| Qwen3.5-122B-A10B | 63.8% | 9.6% |