Sai tops OSWorld 2.0 with 79.38%

by Nate: Microsoft Windows

Simular says sai_borg took the top spot on OSWorld 2.0, scoring a 79.38% partial success rate. That metric tracks the fraction of a task an agent actually completes, not just a clean finish. Sai itself is a coding agent that writes code. It lives on a Windows environment Simular provides. Connect it to Claude Code, Codex, or Cursor through MCP, type out what you want in plain English, and it launches the apps and handles the job on a Windows cloud machine. That setup is genuinely cool to see.