browser-use 1.0.6 — browser-use 1.0.6
- Source
- github.com
- Author
- sarath-menonTop Viber
- Date
Browser Use 1.0.6 changes what the library does when a developer does not choose a model. Agent(task) and ChatBrowserUse() now run bu-4.0 on the Browser Use gateway instead of bu-fast, at $1.80 per million input tokens and $9.00 per million output tokens against $0.12 and $0.60, and the default mode now uses high rather than medium reasoning. The release notes put a number on it: a short test task cost $0.022 instead of $0.0014. Passing llm='bu-fast' keeps the previous model and price, while two new options, bu-4.0-mini and bu-4.0-pro, cover cheap simple tasks and the hardest ones. The release also opens a Browser Use login on the first run without a key and prints the live-view link for cloud browsers. It reverses the direction of 1.0.4, which we covered for halving cost per task through batched screenshot pruning, so a team whose per-task spend jumps after upgrading will find the cause in the default, not in its own code.

Context
Upgrading Browser Use, the open-source Python library that lets an operate a web browser, to 1.0.6 changes which model runs, and what it costs, for anyone who calls Agent(task) or ChatBrowserUse() without naming a model. Since the 1.0 rewrite, that default has been bu-fast, served through Browser Use's gateway, the company's paid service that relays model requests and bills them to one Browser Use key. In 1.0.6 the default becomes BU 4.0 (bu-4.0), according to the release notes. Its cost 15 times as much, $1.80 per million input tokens and $9.00 per million output tokens against $0.12 and $0.60, and it runs at high reasoning effort instead of medium, a setting for how much the model works through each step before acting. The pull request that made the switch notes one more difference: the agent now runs in its default mode rather than the ultrafast mode bu-fast used, which gives the model extra helpers built on the page's accessibility tree.
The maintainers' own figures put the increase per run at roughly 15 to 30 times. One short test task cost $0.022 instead of $0.0014, and the release pull request says full cloud runs average about 30 times more, $0.19 against $0.006, without saying how many runs that average covers. Nothing in the release measures whether BU 4.0 completes more tasks than bu-fast. The switch's pull request reports only that all three BU 4.0 models completed real runs answering one simple question, the year Python was first released, and it does not name the provider model behind them, while bu-fast is documented as gpt-6-luna at medium reasoning. Code that should keep the old behavior has to say so: ='bu-fast' keeps the previous model and price, and the notes suggest llm='bu-4.0-mini' for simple tasks at the lowest cost and llm='bu-4.0-pro' for the hardest. The models documentation the notes cite for prices had not caught up when checked on November 16: it still calls bu-fast the default and the configuration Browser Use evaluates, and lists no BU 4.0 prices.
Two other defaults change. A first run with no Browser Use key on a desktop machine now opens the Browser Use sign-in page and continues once the user approves, instead of stopping with an error; servers, CI jobs, notebooks and piped input still get the missing-key error, and setting BROWSER_USE_AUTO_LOGIN=false turns the new behavior off (#6425). Cloud browsers also now print their live-view link to the error output by default. Anyone with that link can control the browser, so logs that capture it need the care given to a password, and Browser(live_view=False) hides it.
- browser-use 1.0.6 makes bu-4.0 the default model for Agent(task) and ChatBrowserUse() instead of bu-fast, and runs it at high rather than medium reasoning in the default mode.
- The release notes price BU 4.0 at $1.80 input and $9.00 output per 1M tokens, against $0.12 and $0.60 for bu-fast. A short test task cost $0.022 instead of $0.0014.
- Two other tiers are named: bu-4.0-mini for simple tasks at the lowest cost, and bu-4.0-pro for the hardest tasks.
- A first run without an API key now opens a Browser Use login and continues after approval, and cloud browsers print their live-view link.
- AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
Checking sign-in…
Loading comments…


