Calibration
Gert Labs → OSWorld 2.0
OSWorld 2.0 is estimated from Gert Labs with a linear curve fitted on 5 models measured on both: y = 0.7639·x + -0.3853, R² = 0.76, cross-validated error 4.5 pp. It is used for 38 estimates.
| Estimated model | Gert Labs | OSWorld 2.0 | Source |
|---|---|---|---|
| Claude 4 Sonnet | 39.7% | 0.0% | estimated ± 4.5 pp, low confidence |
| DeepSeek V3.2 | 29.6% | 0.0% | estimated ± 4.5 pp, low confidence |
| Gemini 2.5 Pro | 42.0% | 0.0% | estimated ± 4.5 pp, low confidence |
| Gemini 3.1 Flash-Lite | 38.5% | 0.0% | estimated ± 4.5 pp, low confidence |
| Gemini 3.1 Pro | 56.9% | 4.9% | estimated ± 4.5 pp, medium confidence |
| Gemini 3 Flash | 56.6% | 4.7% | estimated ± 4.5 pp, low confidence |
| Gemini 3 Pro | 63.2% | 9.8% | estimated ± 4.5 pp, medium confidence |
| Gemma 4 31B | 35.3% | 0.0% | estimated ± 4.5 pp, low confidence |
| GLM-4.7 | 40.0% | 0.0% | estimated ± 4.5 pp, low confidence |
| GLM-5 | 51.0% | 0.4% | estimated ± 4.5 pp, low confidence |
| GLM-5.1 | 60.1% | 7.4% | estimated ± 4.5 pp, medium confidence |
| GLM-5V-Turbo | 30.8% | 0.0% | estimated ± 4.5 pp, low confidence |
| GPT-4.1 | 25.7% | 0.0% | estimated ± 4.5 pp, low confidence |
| GPT-5.1 | 41.2% | 0.0% | estimated ± 4.5 pp, low confidence |
| GPT-5.1-Codex | 49.7% | 0.0% | estimated ± 4.5 pp, low confidence |
| GPT-5.2-Codex | 51.8% | 1.0% | estimated ± 4.5 pp, low confidence |
| GPT-OSS 120B | 29.6% | 0.0% | estimated ± 4.5 pp, low confidence |
| Grok 4 | 42.3% | 0.0% | estimated ± 4.5 pp, low confidence |
| Grok 4.1 Fast | 47.3% | 0.0% | estimated ± 4.5 pp, low confidence |
| Grok 4.20 | 38.4% | 0.0% | estimated ± 4.5 pp, low confidence |
| Grok 4.3 | 43.9% | 0.0% | estimated ± 4.5 pp, low confidence |
| Grok Build 0.1 | 49.2% | 0.0% | estimated ± 4.5 pp, low confidence |
| Hy3 Preview | 36.9% | 0.0% | estimated ± 4.5 pp, low confidence |
| Kimi K2.5 | 45.9% | 0.0% | estimated ± 4.5 pp, low confidence |
| Kimi K2.5 (Reasoning) | 32.6% | 0.0% | estimated ± 4.5 pp, low confidence |
| MiMo-V2.5 | 46.9% | 0.0% | estimated ± 4.5 pp, low confidence |
| MiMo-V2.5-Pro | 62.7% | 9.4% | estimated ± 4.5 pp, medium confidence |
| MiMo-V2-Pro | 36.7% | 0.0% | estimated ± 4.5 pp, low confidence |
| MiniMax M2.7 | 40.4% | 0.0% | estimated ± 4.5 pp, low confidence |
| Mistral Medium 3.5 128B | 39.1% | 0.0% | estimated ± 4.5 pp, low confidence |
| Qwen3.5 397B | 46.8% | 0.0% | estimated ± 4.5 pp, low confidence |
| Qwen3.6-27B | 54.8% | 3.4% | estimated ± 4.5 pp, low confidence |
| Qwen3.6-35B-A3B | 42.7% | 0.0% | estimated ± 4.5 pp, low confidence |
| Qwen3.6 Plus | 50.6% | 0.1% | estimated ± 4.5 pp, low confidence |
| Qwen3.7 Max | 64.3% | 10.6% | estimated ± 4.5 pp, medium confidence |
| Qwen3 Max | 43.7% | 0.0% | estimated ± 4.5 pp, low confidence |
| Step 3.7 Flash | 51.6% | 0.9% | estimated ± 4.5 pp, low confidence |
| Trinity-Large-Thinking | 32.6% | 0.0% | estimated ± 4.5 pp, low confidence |