Browser Use released version 1.0.6 on November 16, 2026, changing the default model for its Python browser agents from bu-fast to BU 4.0. Existing code that leaves model selection unspecified now uses a model with input and output token rates 15 times higher. The change applies to both Agent(task) and ChatBrowserUse(), according to the release notes.
An upgrade changes the default bill
BU 4.0 costs $1.80 per million input tokens and $9.00 per million output tokens, compared with $0.12 and $0.60 respectively for bu-fast. The default also moves from medium to high reasoning effort.
Those rates do not establish a fixed multiplier for complete tasks: token consumption and the work performed can differ. Browser Use reports that one short test cost $0.022 with the new default versus $0.0014 previously. That is a maintainer-reported example, not an independent measurement or a forecast for every workflow. The release notes explicitly flag the cost increase.
Model selection becomes a budget control
For a team running a recurring browser task, such as collecting supplier prices, upgrading the library can therefore change spending without changing the task prompt. Passing llm='bu-fast' explicitly retains the previous model and its published rates.
The model documentation also lists BU 4.0 Mini at $0.20 per million input tokens and $1.00 per million output tokens, and BU 4.0 Pro at $3.00 and $15.00. Browser Use positions them for simple and difficult tasks respectively; those descriptions are vendor guidance.
The gateway runs in the US. Its result.cost_usd field reports gateway charges for the run and judge, plus any web searches. Developers evaluating the new default should compare completed tasks and recorded costs together. The Browser Use directory profile6.8 out of 10 provides background on the browser-automation tool.
Available now, with different login behavior in CI
Version 1.0.6 is available as an installable Python package requiring Python 3.11 or newer. The package instructions recommend pinning an exact version.
Interactive desktop runs without a key can now open Browser Use login and continue after approval. Servers, CI jobs and piped input retain the missing-key error and login instructions, so unattended deployments still need credentials configured in advance.