Calibration
GDPval-AA → AA Agentic Index
AA Agentic Index is estimated from GDPval-AA with a Hill curve fitted on 72 models measured on both: y = 0.0269 + (1.2000 − 0.0269)·x^1.81 / (0.68360^1.81 + x^1.81), R² = 0.97, cross-validated error 2.8 pp. It is used for 43 estimates.
| Estimated model | GDPval-AA | AA Agentic Index | Source |
|---|---|---|---|
| A.X K2 | 23.2% | 17.2% | estimated ± 2.8 pp, high confidence |
| Apodex 1.1 | 34.8% | 29.4% | estimated ± 2.8 pp, high confidence |
| Apodex 1.1 Mini | 34.8% | 29.4% | estimated ± 2.8 pp, high confidence |
| Claude Opus 5.5 | 68.3% | 61.3% | estimated ± 2.8 pp, medium confidence |
| Claude Sonnet 5.5 | 67.0% | 60.3% | estimated ± 2.8 pp, medium confidence |
| DeepSeek V3 0324 | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| Gemini 4 Argon | 56.3% | 51.1% | estimated ± 2.8 pp, high confidence |
| Gemma 4 12B | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| Gemma 4 26B A4B | 3.4% | 3.2% | estimated ± 2.8 pp, high confidence |
| Gemma 4 E2B | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| Gemma 4 E4B | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| GLM-4.7 | 25.8% | 19.8% | estimated ± 2.8 pp, high confidence |
| GPT-4.1 mini | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| GPT-4.1 nano | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| GPT-4o | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| GPT-4o mini | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| GPT-5.1 | 16.5% | 11.0% | estimated ± 2.8 pp, high confidence |
| GPT-5 (high) | 21.0% | 15.0% | estimated ± 2.8 pp, high confidence |
| GPT-6.1 Sol | 53.8% | 48.8% | estimated ± 2.8 pp, high confidence |
| GPT-6 Luna | 46.9% | 42.1% | estimated ± 2.8 pp, high confidence |
| GPT-6 Sol | 50.5% | 45.6% | estimated ± 2.8 pp, high confidence |
| Granite 4.2 30B | 3.9% | 3.3% | estimated ± 2.8 pp, high confidence |
| Granite 4.2 3B | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| Grok 4.7 | 60.8% | 55.1% | estimated ± 2.8 pp, high confidence |
| K-Exaone | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| Kimi K2.5 | 17.2% | 11.6% | estimated ± 2.8 pp, high confidence |
| Kimi K2.5 (Reasoning) | 17.2% | 11.6% | estimated ± 2.8 pp, high confidence |
| LFM2.5-2.6B | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| Ling 2.6 Flash | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| Ling 3.0 Flash VL | 33.2% | 27.6% | estimated ± 2.8 pp, high confidence |
| Ling 3.0 Tiny | 3.2% | 3.1% | estimated ± 2.8 pp, high confidence |
| Mercury 2.5 | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| MiMo-V2-Flash | 6.6% | 4.4% | estimated ± 2.8 pp, high confidence |
| MiniCPM5-2B | 10.5% | 6.5% | estimated ± 2.8 pp, high confidence |
| Mistral Large 4 | 46.2% | 41.4% | estimated ± 2.8 pp, high confidence |
| Nemotron 3 Nano Omni 30B A3B | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| North Mini Code | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| Qwen3.6 Plus | 24.7% | 18.7% | estimated ± 2.8 pp, high confidence |
| Qwen3.8-Flash-Next | 56.6% | 51.4% | estimated ± 2.8 pp, high confidence |
| Solar Pro 3 | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |
| Solar Pro 4 | 30.5% | 24.7% | estimated ± 2.8 pp, high confidence |
| Step 3.7 Flash | 25.8% | 19.8% | estimated ± 2.8 pp, high confidence |
| Ultravox v0.6 Llama 3.3 70B | 0.0% | 2.7% | estimated ± 2.8 pp, high confidence |