Calibration
OSWorld 2.0 → APEX-Agents-AA
APEX-Agents-AA is estimated from OSWorld 2.0 with a Michaelis–Menten + offset curve fitted on 5 models measured on both: y = 0.0000 + 0.3993·x / (0.01920 + x), R² = 0.91, cross-validated error 2.9 pp. It is used for 18 estimates.
| Estimated model | OSWorld 2.0 | APEX-Agents-AA | Source |
|---|---|---|---|
| Claude Fable 5.1 | 41.7% | 38.2% | estimated ± 2.9 pp, medium confidence |
| Claude Opus 4.7 | 13.9% | 35.1% | estimated ± 2.9 pp, medium confidence |
| Claude Opus 4.7 (Adaptive) | 18.2% | 36.1% | estimated ± 2.9 pp, medium confidence |
| Claude Opus 4.8 | 20.6% | 36.5% | estimated ± 2.9 pp, medium confidence |
| Claude Opus 5 | 70.6% | 38.9% | estimated ± 2.9 pp, low confidence |
| Claude Opus 5.5 | 48.7% | 38.4% | estimated ± 2.9 pp, medium confidence |
| Claude Sonnet 4.6 | 8.3% | 32.4% | estimated ± 2.9 pp, medium confidence |
| Gemini 3.7 Flash | 47.9% | 38.4% | estimated ± 2.9 pp, medium confidence |
| Gemini 3.8 Flash | 59.0% | 38.7% | estimated ± 2.9 pp, low confidence |
| Gemini 4 Argon | 69.2% | 38.9% | estimated ± 2.9 pp, low confidence |
| GPT-5.6 Sol | 62.6% | 38.7% | estimated ± 2.9 pp, low confidence |
| GPT-6 Astra | 72.6% | 38.9% | estimated ± 2.9 pp, low confidence |
| GPT-6 Sol | 60.5% | 38.7% | estimated ± 2.9 pp, low confidence |
| MiniMax M3 | 4.6% | 28.2% | estimated ± 2.9 pp, medium confidence |
| Muse Spark 1.1 | 14.2% | 35.2% | estimated ± 2.9 pp, medium confidence |
| Muse Spark 1.3 | 66.9% | 38.8% | estimated ± 2.9 pp, low confidence |
| Qwen3.8-Flash-Next | 19.4% | 36.3% | estimated ± 2.9 pp, medium confidence |
| Qwen3.8 Max | 19.4% | 36.3% | estimated ± 2.9 pp, medium confidence |