Cua Driver Bench: 58% pass rate on background macOS desktop tasks vs Codex 33%
TianbaoX · x · 2026-11-07
Cua launched Driver Bench with Scale AI's AgentEnv, testing whether agents can complete desktop tasks in the background while the user keeps working. On macOS, Cua Driver fully passed 58% of tasks, versus 33% for Codex computer use, 20% for Vercel's agent-desktop, and just 3% for browser-use — a rare head-to-head on background desktop automation.
More from coding & agent
- HackerRank CEO stress-tests his own AI interviewer to show how easily it can be cheated — emeka_boris · 2026-11-07
- Nous launches Portal: one account for 200+ models, bundled tools, and agent cloud hosting — NousResearch · 2026-11-07
- Playing Factorio Made Me Better at Managing AI Agents — billyjhowell · 2026-11-07
- LangChain's deepagents Adds Browser Use and Async Subagents — LangChain · 2026-11-07
- DepthFirst shares its secure cloud dev environment for coding agents, built in one week — andreamichi · 2026-11-07
- BoopFace: MIT-licensed SVG robot avatar Web Component with an ESP32/STM32 embedded engine — Additional-Ad5527 · 2026-11-07