Calibration
GDPval-AA → OSWorld-Verified
OSWorld-Verified is estimated from GDPval-AA with a linear curve fitted on 20 models measured on both: y = 0.5249·x + 0.5555, R² = 0.46, cross-validated error 8.7 pp. It is used for 51 estimates.
| Estimated model | GDPval-AA | OSWorld-Verified | Source |
|---|---|---|---|
| A.X K2 | 23.2% | 67.7% | estimated ± 8.7 pp, low confidence |
| Apodex 1.1 Mini | 34.8% | 73.8% | estimated ± 8.7 pp, low confidence |
| Celeris-1 | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Claude Sonnet 5.5 | 67.0% | 90.7% | estimated ± 8.7 pp, low confidence |
| Command A+ | 0.9% | 56.0% | estimated ± 8.7 pp, low confidence |
| DeepSeek V3 | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| DeepSeek V3 0324 | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Gemma 3 27B | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Gemma 4 12B | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Gemma 4 26B A4B | 3.4% | 57.3% | estimated ± 8.7 pp, low confidence |
| Gemma 4 E2B | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Gemma 4 E4B | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| GLM-5.2 | 43.7% | 78.5% | estimated ± 8.7 pp, low confidence |
| GPT-4.1 mini | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| GPT-4.1 nano | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| GPT-4o | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| GPT-4o mini | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| GPT-6.1 Sol | 53.8% | 83.8% | estimated ± 8.7 pp, low confidence |
| GPT-6 Luna | 46.9% | 80.2% | estimated ± 8.7 pp, low confidence |
| GPT-OSS 20B | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Granite 4.2 30B | 3.9% | 57.6% | estimated ± 8.7 pp, low confidence |
| Granite 4.2 3B | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Granite 4.2 8B | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Grok 4.5 | 44.5% | 78.9% | estimated ± 8.7 pp, low confidence |
| Grok 4.7 | 60.8% | 87.5% | estimated ± 8.7 pp, low confidence |
| Hy3 | 28.3% | 70.4% | estimated ± 8.7 pp, low confidence |
| K-Exaone | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Kimi K2.7 Code | 27.0% | 69.7% | estimated ± 8.7 pp, low confidence |
| LFM2.5-2.6B | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Ling 2.6 Flash | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Ling 3.0 Flash FP8 | 22.4% | 67.3% | estimated ± 8.7 pp, low confidence |
| Ling 3.0 Flash VL | 33.2% | 73.0% | estimated ± 8.7 pp, low confidence |
| Ling 3.0 Tiny | 3.2% | 57.2% | estimated ± 8.7 pp, low confidence |
| Llama 4 Maverick | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Llama 4 Scout | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| MiMo-V2-Flash | 6.6% | 59.0% | estimated ± 8.7 pp, low confidence |
| MiniCPM5-2B | 10.5% | 61.1% | estimated ± 8.7 pp, low confidence |
| Mistral Large 3 | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Mistral Large 4 | 46.2% | 79.8% | estimated ± 8.7 pp, low confidence |
| Mistral Small 4 | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Mistral Small 4 (Reasoning) | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Muse Spark 1.2 | 48.9% | 81.2% | estimated ± 8.7 pp, low confidence |
| Nemotron 3 Nano 30B | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Nemotron 3 Nano Omni 30B A3B | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Nemotron 3 Super 100B | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| North Mini Code | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Quasar 438B | 34.2% | 73.5% | estimated ± 8.7 pp, low confidence |
| Qwen3.8 Max Preview | 58.6% | 86.3% | estimated ± 8.7 pp, low confidence |
| Solar Pro 3 | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Trinity-Large-Preview | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |
| Ultravox v0.6 Llama 3.3 70B | 0.0% | 55.6% | estimated ± 8.7 pp, low confidence |