Browser Use 1.0.0rc4 Released with Rewritten Python Engine
Open-source browser automation project Browser Use released version 1.0.0rc4 on Nov 9, 2026, featuring a rewritten Python engine and lean codebase.
Stock photo for illustration only, not from the actual event
- Browser Use published release candidate 1.0.0rc4 on November 9, 2026.
- The engine was rewritten so agents write Python in a REPL instead of picking single actions.
- Codebase shrunk from 68,815 to 13,154 lines while preserving core 0.x APIs.
- Scored 62.6 on BU Bench v2, running 18% cheaper and 15% faster than predecessors.
On November 9, 2026, the open-source browser automation project GitHub browser-use published release candidate 1.0.0rc4 as a pre-release on GitHub, marking the first 1.0 candidate that engineering teams can actually pull and evaluate ahead of the stable rollout. The repository currently accumulates 120,445 stars and 13,459 forks.
The new build replaces the core engine entirely, according to the official release notes. Instead of choosing a single action per step from a DOM dump, the agent now writes Python inside one persistent REPL session, leveraging browser-harness helpers and raw CDP within the same session, all built on top of Pydantic AI.
Stock photo for illustration only, not from the actual event
As a result of this architectural overhaul, the codebase shrank dramatically from 68,815 down to 13,154 lines of Python code. Meanwhile, the essential 0.x API suite—including Agent, Browser, Tools, run(), and final_result()—remains fully preserved to ensure seamless backwards compatibility for existing deployments.
Stock photo for illustration only, not from the actual event
New defaults alter post-installation behavior significantly. Setting llm=None now routes execution directly to the bu-fast model via the Browser Use gateway, while instantiating Browser() spins up a Browser Use Cloud browser unless developers explicitly pass Browser('chromium') for a local instance. Furthermore, run() caps max_steps at 100 instead of 500 with a 1,700-second timeout, a single BROWSER_USE_API_KEY replaces provider-specific keys, and Python 3.11 or newer is strictly required.
Shifting browser agents toward direct Python code generation via a persistent REPL represents a major leap in flexibility, moving away from rigid step-by-step action selection. Slicing the codebase by nearly 80 percent also drastically simplifies security auditing and supply-chain management for enterprise engineering teams.
Evaluated on BU Bench v2—consisting of 60 tasks across 3 reps on cloud browsers—the 1.0 ultrafast mode scored 62.6 at a cost of 0.071 dollars per task and a runtime of 585 seconds per task. While performance differences against Browser Use Pi are reported as not significant, the new build achieves an 18 percent cost reduction and a 15 percent speedup.
Version 1.0 also introduces typed results through output_model_schema, allowing result.structured_output to return a fully validated object. Additional features include file upload and download capabilities, secret redaction via type_secret(), domain restrictions including block_ip_addresses, a web_search=True flag, an ultrafast mode powered by accessibility-tree IDs, and strict dependency pinning with == as a supply-chain security policy.
Regarding rollout logistics, pre-releases rc1 through rc3 were never published to PyPI, meaning a standard pip install browser-use will continue installing version 0.13.14 until 1.0.0 officially ships. Engineering teams wishing to adopt the rewritten engine immediately must install version 1.0.0rc4 directly from the release tag.
Source: Dev.to
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment