AI NewsDev toolsAnnouncement

browser-use 1.0.6 makes BU 4.0 the default at 15x the token price

browser-use 1.0.6 makes bu-4.0 the default model, and it says the new default costs 15 times more per token.

AI News

Editorial2 min read

LinkedInX
browser-use project header on GitHub

Why it mattersA team using browser-use without picking a model now spends 15 times more per task. The release says passing the old model keeps the old price.

A default model change in a library can cost a team 15 times more on the next run. browser-use published release 1.0.6 of its browser agent on GitHub. It makes bu-4.0 the default. The release page says bu-4.0 tokens cost 15 times as much as the old default, bu-fast.

What changed in the default

In the old version, Agent(task) and ChatBrowserUse() ran bu-fast on the Browser Use gateway when the caller passed no model. In 1.0.6 the same two calls run bu-4.0. The release page says bu-4.0 also runs at high reasoning in the default mode, where bu-fast ran at medium. A caller who wants the old model and the old price must pass llm='bu-fast'.

The release page lists two other model names. bu-4.0-mini is the option it names for simple tasks at the lowest cost. bu-4.0-pro is the option it names for the hardest tasks.

The price the release page shows

The release page states the prices per 1 million tokens. bu-fast is $0.12 for input and $0.60 for output. bu-4.0 is $1.80 for input and $9.00 for output. That is 15 times each number.

The release page then shows one worked example. A short test task cost $0.0014 on bu-fast. The same task cost $0.022 on bu-4.0. That is close to 16 times. The release does not say which task or how many steps it took.

Two other changes in the same release

The release page says the first run without an API key now opens the Browser Use login flow. The agent then resumes once the login is approved. Before 1.0.6, a run without a key failed.

Cloud browsers now print a live-view link so the caller can watch the run in a browser tab. The release also says "Judge gotchas stay grounded in what the run observed". Judge is the part that decides whether a task has succeeded.

The release lists four pull requests: #6430, #6425, #6418 and #6431. It was cut by GitHub user sarath-menon.

The cost figures and the worked example come from browser-use's own release page. A team's own cost on a real workload may differ. The gain in capability from bu-4.0 over bu-fast is not reported in this release.

A team running browser-use in production has one thing to check before upgrading. Read the model passed to Agent and ChatBrowserUse. If it is not set, the next run pays the new price. Setting it back to bu-fast is the release page's own answer. A team that wants bu-4.0 only on the hardest tasks can route by work type between bu-4.0-mini, bu-4.0 and bu-4.0-pro.

Source

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.

Share
LinkedInX
Start a project