
Browser Use 1.0.4 changes how its browser agent removes old screenshots from a conversation: the project now batches pruning after 20 tool results contain screenshots, instead of removing one on each new screenshot. Its November 13 release notes say repeated edits to earlier messages were disrupting provider prompt caching during screenshot-heavy tasks.
Prompt caching can reduce repeated processing when a provider recognizes an unchanged prefix, but the savings depend on the provider and the workload. In its own 200-task BU Bench comparison, the project reports model cost per task of $0.208 versus $0.434 for version 1.0.3, with similar scores. Those figures are project-reported, not independent measurements or a guarantee for other workflows.
Other changes
The release says callers using BU_NAME receive their own Chromium when Chrome cannot be used. It also adds upgrade guidance to the command-line interface and Model Context Protocol server when a browser setup error may have been fixed in a later version.
Browser Use is an MIT-licensed Python browser-agent project with local and hosted options. Its OpenSourceCoder.com repository profile links to the upstream source and documentation. A browser agent’s permissions, chosen model provider and any hosted service charges remain separate decisions from the project’s code license.
Review access before automating
Agents can encounter account sessions, page content and actions that affect real systems. Review what the browser can access, use appropriate isolation and inspect outputs before applying changes. The project documentation notes that domain restrictions are not a general sandbox.
We have reviewed the release notes and repository documentation but have not installed or benchmarked this version.
EXPLORE THE SOURCE
Projects in this story
GitHub repositories
- browser-use/browser-use ↗View project details
Original reporting. Primary sources reviewed November 14, 2026; no independent hands-on test claimed.



