OSWorld 2.0 - reported binary completion, patch and step limit unspecified
Agents
Strict completion on OSWorld 2.0 when the release omits dated task patch and step limit. Source-dependent averaging is preserved without inferring the 108-task, 500-step protocol.
Top models (higher is better)
| Model | Score |
|---|---|
| Qwen3.8-Flash-Next | 19.4 |