Calibration
GDPval-AA → DeepSearchQA
DeepSearchQA is estimated from GDPval-AA with a offset logistic curve fitted on 14 models measured on both: y = 0.0000 + (0.9055 − 0.0000) / (1 + exp(−12.62·(x − 0.0386))), R² = 0.85, cross-validated error 8.4 pp. It is used for 40 estimates.
| Estimated model | GDPval-AA | DeepSearchQA | Source |
|---|---|---|---|
| A.X K2 | 23.2% | 83.3% | estimated ± 8.4 pp, medium confidence |
| Apodex 1.1 Mini | 34.8% | 88.8% | estimated ± 8.4 pp, medium confidence |
| Claude Opus 5.5 | 68.3% | 90.5% | estimated ± 8.4 pp, low confidence |
| Claude Sonnet 5.5 | 67.0% | 90.5% | estimated ± 8.4 pp, low confidence |
| DeepSeek V3 0324 | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| DeepSeek V4.1 Flash | 55.0% | 90.4% | estimated ± 8.4 pp, medium confidence |
| Gemini 4 Argon | 56.3% | 90.4% | estimated ± 8.4 pp, medium confidence |
| Gemma 4 12B | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| Gemma 4 26B A4B | 3.4% | 44.0% | estimated ± 8.4 pp, medium confidence |
| Gemma 4 E2B | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| Gemma 4 E4B | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| GPT-4.1 mini | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| GPT-4.1 nano | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| GPT-4o | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| GPT-4o mini | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| GPT-5.1 | 16.5% | 75.3% | estimated ± 8.4 pp, medium confidence |
| GPT-5 (high) | 21.0% | 81.2% | estimated ± 8.4 pp, medium confidence |
| GPT-6.1 Sol | 53.8% | 90.4% | estimated ± 8.4 pp, medium confidence |
| GPT-6 Luna | 46.9% | 90.2% | estimated ± 8.4 pp, medium confidence |
| GPT-6 Sol | 50.5% | 90.3% | estimated ± 8.4 pp, medium confidence |
| Granite 4.2 30B | 3.9% | 45.4% | estimated ± 8.4 pp, medium confidence |
| Granite 4.2 3B | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| Grok 4.7 | 60.8% | 90.5% | estimated ± 8.4 pp, medium confidence |
| K-Exaone | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| LFM2.5-2.6B | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| Ling 2.6 Flash | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| Ling 3.0 Flash VL | 33.2% | 88.4% | estimated ± 8.4 pp, medium confidence |
| Ling 3.0 Tiny | 3.2% | 43.4% | estimated ± 8.4 pp, medium confidence |
| Ling 3.1 Flash | 56.1% | 90.4% | estimated ± 8.4 pp, medium confidence |
| MiMo-V2.6-Flash | 55.5% | 90.4% | estimated ± 8.4 pp, medium confidence |
| MiMo-V2.6-Pro | 59.3% | 90.5% | estimated ± 8.4 pp, medium confidence |
| MiMo-V2-Flash | 6.6% | 53.0% | estimated ± 8.4 pp, medium confidence |
| MiniCPM5-2B | 10.5% | 63.2% | estimated ± 8.4 pp, medium confidence |
| Mistral Large 4 | 46.2% | 90.1% | estimated ± 8.4 pp, medium confidence |
| Nemotron 3 Nano Omni 30B A3B | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| North Mini Code | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| Qwen3.6 Plus | 24.7% | 84.5% | estimated ± 8.4 pp, medium confidence |
| Qwen3.8-Flash-Next | 56.6% | 90.4% | estimated ± 8.4 pp, medium confidence |
| Solar Pro 3 | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |
| Ultravox v0.6 Llama 3.3 70B | 0.0% | 34.5% | estimated ± 8.4 pp, medium confidence |