Co-Arena: live arena for computer-use agents hits 55K steps in 6 days
Scobleizer · x · 2026-08-12
Co-Arena, a live arena and advanced benchmark for computer-use agents (CUA), has accumulated over 55,000 steps in six days. It runs two frontier models side-by-side in identical sandboxes, recording every step, with human blind judging. It aims to address benchmark saturation by using real tasks from real users.
More from coding & agent
- Anthropic Engineer: Stop Prompting Claude, Build Systems That Prompt Themselves — Fowe · 2026-08-12
- AI Agent Preferences Reshape Dev Ecosystem: Drizzle Surpasses Prisma in NPM Downloads — tristanbob · 2026-08-12
- Developer Shares Practice of Using GitHub Copilot for Multi-Model Code Reviews — DanWahlin · 2026-08-12
- Expert Warns AI Watermarks Could Increase Code Entropy, Harming Coding Agents — wightmanr · 2026-08-12
- Claude Code Skills Catalog Adds NTM Multi-Agent Orchestration Skills, Downloads Exceed 80K — doodlestein · 2026-08-12
- Local Agent Memory Bottleneck: Vector Search Fails, Structured Data Models Emerge — CopyPasteVeteran · 2026-08-12