# OSWorld 2.0 → ExploitGym

Calibration · https://benchgap.net/calibration/osworld2/current/to/exploitgym/current · data retrieved 2026-11-18

ExploitGym is estimated from OSWorld 2.0 with a Michaelis–Menten curve fitted on 10 models measured on both: y = 1.1202·x / (1.78710 + x), R² = 0.55, cross-validated error 10.4 pp. It is used for 2 estimates.

| Estimated model | OSWorld 2.0 | ExploitGym | Source |
|---|---|---|---|
| [Qwen3.8-Flash-Next](https://benchgap.net/model/qwen3-8-flash-next.md) | 19.4% | 11.0% | estimated ± 10.4 pp, low confidence |
| [Qwen3.7 Plus](https://benchgap.net/model/qwen3-7-plus.md) | 2.8% | 1.7% | estimated ± 10.4 pp, low confidence |

## Fitted on

| Model | OSWorld 2.0 | ExploitGym |
|---|---|---|
| [Claude Opus 5.5](https://benchgap.net/model/claude-opus-5-5.md) | 48.7% | 33.3% |
| [Gemini 4 Argon](https://benchgap.net/model/gemini-4-argon.md) | 69.2% | 39.5% |
| [GPT-5.5](https://benchgap.net/model/gpt-5-5.md) | 13.0% | 13.4% |
| [GPT-5.6 Luna](https://benchgap.net/model/gpt-5-6-luna.md) | 45.6% | 12.4% |
| [GPT-5.6 Sol](https://benchgap.net/model/gpt-5-6-sol.md) | 62.6% | 33.7% |
| [GPT-5.6 Terra](https://benchgap.net/model/gpt-5-6-terra.md) | 50.2% | 23.2% |
| [GPT-6 Astra](https://benchgap.net/model/gpt-6-astra.md) | 72.6% | 42.4% |
| [GPT-6 Sol](https://benchgap.net/model/gpt-6-sol.md) | 60.5% | 22.1% |
| [Kimi K3.1 Pro](https://benchgap.net/model/kimi-3-1-pro.md) | 79.6% | 20.7% |
| [Muse Spark 1.1](https://benchgap.net/model/muse-spark-1-1.md) | 14.2% | 0.8% |
