Intel / post

Cua Adds a Guide for Running OSWorld Benchmarks on Its Fleets

Key takeaways · AI-distilled

The recipe starts from the official OSWorld Ubuntu image and its applications, adds Cua Driver, publishes the VM disk to a registry, and claims a desktop through the SDK.
Task setup and scoring use OSWorld's own setup tools and evaluators against the Fleet desktop, with Cua Driver handling actions. Because Fleet does not revert snapshots, each task needs a freshly claimed desktop.
The guide includes a Python sample that claims the desktop, takes a screenshot and calls Cua Driver. It requires Fleet credentials, a registry image and cua-sandbox installed from main.

Terms in this piece · Glossary

benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
sandbox — An isolated environment where AI-generated code or agent actions run without being able to touch anything real.

Why it matters

Lowers the setup cost for running the standard OSWorld computer-use against agents built on Cua's stack.