Browser Use deletes 90% of its code, hands agent raw CDP
The team's own numbers say the thinner harness wins, and nobody outside the team has posted a run yet.
Browser Use shipped 1.0.0 and deleted 90% of the old library. The predefined state and action space is gone. In its place the agent writes Python in a REPL against raw CDP, the Chrome DevTools Protocol that drives the browser directly.
Predefined state and action space make it worse.
The headline claim from the project account is that this is not a tradeoff. Same models, up to 50% better, 2-5x cheaper, 2-3x faster, and what the post calls a SOTA web agent. Gregor Zunic posted his own version of the launch with slightly rounder figures, 50% better, 5x cheaper, 3x faster, and the one line summary that explains the whole release.
The argument
The case being made is that the scaffolding was the problem. A curated action list and a structured page state are a translation layer between the model and the browser, and the team's position is that the translation hurts. Strip it out, let the model write code against the protocol, and the same models do better on the same work.
That is a strong claim about harness design generally, not just about this library. It is also a claim, not an independent result.
For people building on it
Every number in this launch comes from the people who shipped it. There is no outside run posted, no named benchmark in the announcement, and no breakdown of which tasks produced the up to 50% figure or what the 2-5x cheaper range was measured against. The spread between the project account's numbers and Zunic's own post is itself a hint that these are rounded summaries rather than a published eval. If you are deciding whether to migrate an existing agent, budget for your own before and after on your own task set.
The other practical note is in the last line of the announcement. The team says it ships fast from now on and asks people to pin with ==, which is a reasonable thing to do when a project has just removed most of its surface area in one release.
Simple harness wins.
Know what Jev is good at before you bet on it
One email a day. Under three minutes. What builders found out about Jev yesterday.

