Source: github.com/browser-use/browser-use/releases/tag/1.0.4rc1
The release candidate for version 1.0.4 changes how the agent manages screenshots. Old screenshots are now dropped in one batch once 20 tool results carry one, rather than on every new screenshot. The maintainers say the earlier approach rewrote an earlier message on each prune, which broke the provider's prompt cache.
The cache break had a measurable cost. In a BU Bench 1.0.3 run, one 705-step task read 27.8 million input tokens, of which 1.6 million came from cache, so screenshot-heavy runs paid full input price on almost every step.
Two other changes are included. Callers that set BU_NAME now get their own Chromium when Chrome cannot be used. When a browser setup error may be fixed in a newer release, the CLI and MCP now tell the agent to upgrade.
Why it matters if you build agents
Developers running screenshot-heavy browser agents should see fewer uncached input tokens, because earlier prompt content is no longer rewritten on each prune. Teams that depend on BU_NAME should check whether their runs now start a separate Chromium. Because this is a release candidate, it is installed with pip install 'browser-use==1.0.4rc1' and should be tested before the final 1.0.4 ships.
Get stories like this every weekday
One short email with the day's agent tooling news.