Skip to main content

From the Zalos team

Engineering notes on building agents that work with the tools you already use.

Engineering

A better browser agent, without a new browser

How we built Zalos’s harness and reached 94/100 on BU-Bench-V1 using GPT-5.5 with Chrome.

Read the article

Reported BU-Bench-V1 scores: Zalos 94%, Aside 93%, Claude Code 80%, Browser Use OSS 80%, and Browser Use Cloud 78%. Separate evaluation conditions differ.