Calibration
GDPval-AA → Toolathlon
Toolathlon is estimated from GDPval-AA with a offset logistic curve fitted on 19 models measured on both: y = 0.2691 + (0.5585 − 0.2691) / (1 + exp(−58.59·(x − 0.2463))), R² = 0.74, cross-validated error 6.3 pp. It is used for 73 estimates.
| Estimated model | GDPval-AA | Toolathlon | Source |
|---|---|---|---|
| A.X K2 | 23.2% | 35.6% | estimated ± 6.3 pp, medium confidence |
| Apodex 1.1 | 34.8% | 55.8% | estimated ± 6.3 pp, medium confidence |
| Apodex 1.1 Mini | 34.8% | 55.8% | estimated ± 6.3 pp, medium confidence |
| Celeris-1 | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Claude Opus 5.5 | 68.3% | 55.9% | estimated ± 6.3 pp, low confidence |
| Claude Sonnet 5.5 | 67.0% | 55.9% | estimated ± 6.3 pp, low confidence |
| Command A+ | 0.9% | 26.9% | estimated ± 6.3 pp, low confidence |
| DeepSeek V3 | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| DeepSeek V3 0324 | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| DeepSeek V4.1 Flash | 55.0% | 55.9% | estimated ± 6.3 pp, medium confidence |
| Gemini 2.5 Pro | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Gemini 3.1 Pro | 14.7% | 27.0% | estimated ± 6.3 pp, low confidence |
| Gemini 3.5 Flash-Lite | 24.3% | 40.0% | estimated ± 6.3 pp, medium confidence |
| Gemini 3.6 Flash | 39.3% | 55.8% | estimated ± 6.3 pp, medium confidence |
| Gemini 4 Argon | 56.3% | 55.9% | estimated ± 6.3 pp, low confidence |
| Gemma 3 27B | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Gemma 4 12B | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Gemma 4 26B A4B | 3.4% | 26.9% | estimated ± 6.3 pp, low confidence |
| Gemma 4 31B | 6.1% | 26.9% | estimated ± 6.3 pp, low confidence |
| Gemma 4 E2B | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Gemma 4 E4B | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| GLM-5.3 | 57.6% | 55.9% | estimated ± 6.3 pp, low confidence |
| GPT-4.1 mini | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| GPT-4.1 nano | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| GPT-4o | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| GPT-4o mini | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| GPT-5.1 | 16.5% | 27.2% | estimated ± 6.3 pp, low confidence |
| GPT-5 (high) | 21.0% | 30.0% | estimated ± 6.3 pp, medium confidence |
| GPT-6.1 Sol | 53.8% | 55.9% | estimated ± 6.3 pp, medium confidence |
| GPT-6 Luna | 46.9% | 55.9% | estimated ± 6.3 pp, medium confidence |
| GPT-6 Sol | 50.5% | 55.9% | estimated ± 6.3 pp, medium confidence |
| GPT-OSS 120B | 5.6% | 26.9% | estimated ± 6.3 pp, low confidence |
| GPT-OSS 20B | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Grok 4.3 | 29.2% | 54.0% | estimated ± 6.3 pp, medium confidence |
| Grok 4.5 | 44.5% | 55.9% | estimated ± 6.3 pp, medium confidence |
| Grok 4.7 | 60.8% | 55.9% | estimated ± 6.3 pp, low confidence |
| Hy3 | 28.3% | 52.8% | estimated ± 6.3 pp, medium confidence |
| Hy3 Preview | 35.8% | 55.8% | estimated ± 6.3 pp, medium confidence |
| K-Exaone | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Kimi K2.7 Code | 27.0% | 50.1% | estimated ± 6.3 pp, medium confidence |
| Ling 2.6 Flash | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Ling 3.0 Flash FP8 | 22.4% | 33.1% | estimated ± 6.3 pp, medium confidence |
| Ling 3.0 Flash VL | 33.2% | 55.7% | estimated ± 6.3 pp, medium confidence |
| Ling 3.0 Tiny | 3.2% | 26.9% | estimated ± 6.3 pp, low confidence |
| Ling 3.1 Flash | 56.1% | 55.9% | estimated ± 6.3 pp, low confidence |
| Llama 4 Maverick | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Llama 4 Scout | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| MiMo-V2.6-Flash | 55.5% | 55.9% | estimated ± 6.3 pp, medium confidence |
| MiMo-V2.6-Pro | 59.3% | 55.9% | estimated ± 6.3 pp, low confidence |
| MiMo-V2-Flash | 6.6% | 26.9% | estimated ± 6.3 pp, low confidence |
| MiniCPM5-2B | 10.5% | 26.9% | estimated ± 6.3 pp, low confidence |
| Mistral Large 3 | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Mistral Large 4 | 46.2% | 55.9% | estimated ± 6.3 pp, medium confidence |
| Mistral Small 4 | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Mistral Small 4 (Reasoning) | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Muse Glimmer 30B | 14.5% | 27.0% | estimated ± 6.3 pp, low confidence |
| Muse Spark | 25.1% | 43.3% | estimated ± 6.3 pp, medium confidence |
| Muse Spark 1.2 | 48.9% | 55.9% | estimated ± 6.3 pp, medium confidence |
| Nemotron 3 Nano 30B | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Nemotron 3 Nano Omni 30B A3B | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Nemotron 3 Super 100B | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| North Mini Code | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Quasar 438B | 34.2% | 55.7% | estimated ± 6.3 pp, medium confidence |
| Qwen3.6-27B | 24.4% | 40.4% | estimated ± 6.3 pp, medium confidence |
| Qwen3.7 Max | 31.6% | 55.4% | estimated ± 6.3 pp, medium confidence |
| Qwen3.7 Plus | 13.5% | 27.0% | estimated ± 6.3 pp, low confidence |
| Qwen3.8-27B | 46.2% | 55.9% | estimated ± 6.3 pp, medium confidence |
| Qwen3.8-Flash-Next | 56.6% | 55.9% | estimated ± 6.3 pp, low confidence |
| Qwen3.8 Max Preview | 58.6% | 55.9% | estimated ± 6.3 pp, low confidence |
| Solar Pro 3 | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Trinity-Large-Preview | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Trinity-Large-Thinking | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |
| Ultravox v0.6 Llama 3.3 70B | 0.0% | 26.9% | estimated ± 6.3 pp, low confidence |