Fig Releases RealCUA: 600+ Hours of Real Open-Ended Computer-Use Work Data
vsikka · x · 2026-11-06
Fig.inc released RealCUA v0.1, billed as the first computer-use dataset of open-ended work in evolving environments, targeting the frontier-model challenge of deriving learning signals from open-ended knowledge work.
Key specs:
- 600+ hours (growing) of real multi-objective trajectories with 27,000+ app transitions, longest continuous trajectory of 12 hours, 87% of hours from 1h+ recordings
- Covers 35 apps: browsers, Slack/Mail, Cursor/Terminal, Notion/Figma and more
- Interwoven workflows: multiple independent goals sharing on-screen context across thousands of steps, with real starting states and checkable outcomes
- 30fps H.264 video, sub-millisecond input alignment, step-level JSON actions, accessibility-tree/DOM state, and trajectory segmentation into workflows and atomic tasks
An 11-hour sample and a trajectory viewer are available; academic and open-source groups can request access.
More from coding & agent
- GitHub's new Git storage handles 7,200 pushes/sec from 5,000 concurrent agents — arpit_bhayani · 2026-11-06
- GitHub rebuilds Git storage for thousands of concurrent AI agents — arpit_bhayani · 2026-11-06
- OpenAI built Pando for Codex: 5x lower Git push latency, ~1PB transferred daily — sven_ai · 2026-11-06
- Snorkel AI: agents peek at reward code ~25% of the time rather than asking users — ajratner · 2026-11-06
- Subagents fail because summaries are lossy — checkpoint agents fork context instead — narphorium · 2026-11-06
- Building a pure-code lightbulb with Claude: WebGL shader, MediaPipe tracking and self-made test videos — ramramlab · 2026-11-06