Reported News
Browser Use 1.0 deletes 80% of the library: the agent now writes Python straight on Chrome DevTools Protocol
browser-use 1.0.0, released on GitHub and PyPI on November 10, 2026 (UTC), is a rewrite that removes 136,866 lines: "The agent now writes Python in one persistent REPL, straight on Chrome DevTools Protocol." By default it now uses a Browser Use cloud browser and the bu-fast model on the Browser Use gateway.
- Changes reported
- 6
- listed below, each from a source we read
- We tested
- Not tested yet
- reported from sources; we did not run it ourselves
- Gaps we flag
- 4
- things we could not verify
- Sources
- 2
- 1 primary, linked below
What changed
| # | What the source says | Status |
|---|---|---|
| 01 | The architecture is new: the model writes Python cells that run in one persistent REPL directly on Chrome DevTools Protocol, with no DOM serializer or action schema in between. The release notes say 136,866 lines (80% of the library) were deleted. | Reported |
| 02 | Defaults change: Browser() is now a Browser Use Cloud browser (Browser('chromium') for a local one), llm=None uses bu-fast on the Browser Use gateway, and one BROWSER_USE_API_KEY replaces per-provider keys. Your own OpenAI, Anthropic, Google or Pydantic AI model still works. | Reported |
| 03 | run() now defaults to max_steps=100 (was 500) and a 1,700-second timeout, and after configuration checks pass it returns status='error' instead of raising on key, model or browser-start failures. | Reported |
| 04 | Breaking changes are explicit: unsupported 0.x options raise a TypeError before anything is billed, removed imports raise an ImportError naming the replacement, LangChain chat models are rejected, the old bu-2-0 and bu-1-0 gateway models are refused, and the bu and browser console scripts are gone. | Reported |
| 05 | New in 1.0: typed results via output_model_schema, uploads and downloads in a workspace folder, redacted secrets with type_secret(), web search billed to the Browser Use key, an MCP server via --cli-mcp, and Docker images for 1.0.0. | Reported |
| 06 | Every dependency is now pinned exactly with a committed lockfile, which the project lists as a known issue: projects that need a different pydantic, mcp, google-genai or pillow version cannot install 1.0 alongside it. | Reported |
reported by the sourcemeasured or run by usnot verified
Primary source: GitHub release: browser-use/browser-use 1.0.0. We are the fast, independent index; the vendor’s own page is the source of record for this release.
Who should act, and who can wait
Anyone running browser-use 0.x in production should pin before upgrading: the project itself says to use pip install 'browser-use<1' if you need 0.x, and to read MIGRATING.md first. The biggest silent-looking change is that a plain Browser() now starts a paid Browser Use Cloud browser, not local Chromium, so scripts that hit localhost, staging on a private IP or an internal tool need Browser('chromium').
Teams that already bring their own model key and keep Browser('chromium') explicit should mostly hit loud, early errors rather than behaviour drift, because 1.0 raises on options it cannot honour. Projects built on LangChain chat models, the bu or browser commands, or the removed extras (aws, oci, video, eval and others) have real porting work.
What the benchmark actually shows
The headline claim is up to 33 points higher, 2.7 to 5.2x cheaper and 2 to 3x faster with the same models. The one benchmark table in the notes is narrower: on BU Bench v2 (60 tasks x 3 reps, run 2026-11-09), 1.0 in ultrafast mode scored 62.5 versus 56.3 for Browser Use Pi, at $0.070 versus $0.082 per task.
The project is candid that an OpenAI outage hit the run: with provider errors excluded, the paired score difference was +1.1 points with an interval of -3.2 to +5.7, which it calls no significant difference. Our reading: the defensible claim from this table is same quality, somewhat cheaper and faster than Pi, not a large accuracy jump. Both arms used gpt-6-luna, and the default bu-fast model was not part of the run.
| Agent | Score [95% CI] | $/task | s/task |
|---|---|---|---|
| browser-use 1.0 (ultrafast) | 62.5 [58.3, 66.5] | 0.070 | 650 |
| Browser Use Pi (e7007f2) | 56.3 [50.2, 62.4] | 0.082 | 743 |
The question the release leaves open
1.0 moves the zero-config path onto Browser Use's own cloud browser and model gateway, billed to one Browser Use key. The notes also say result.cost_usd covers model and web search charges but not the cloud browser or its proxy, so the cost a run reports is not the full bill. Builders comparing open-source browser agents should price the default path, not just the library.
In the source’s words
“The agent now writes Python in one persistent REPL, straight on Chrome DevTools Protocol.”
“So the raw gap overstates quality.”
“Tasks on localhost or private IPs need”
What we have not verified
- We did not run this release ourselves; all benchmark figures are the project's own and we did not reproduce BU Bench v2.
- The release notes do not give Browser Use Cloud browser pricing, so we could not say what the new default Browser() costs per run.
- We did not check MIGRATING.md line by line or test how many real 0.x scripts upgrade without changes.
- The 'up to 33 points higher' and '2.7-5.2x cheaper' headline figures are not backed by a table in the release notes beyond the Pi comparison; we did not open the linked benchmarks page.
This brief is Reported: written from the sources below, not from running the release ourselves.
Sources
- GitHub release: browser-use/browser-use 1.0.0primary source · fetched 2026-11-10
- PyPI: browser-use 1.0.0source · fetched 2026-11-10
Related
Spotted an error? See our corrections policy or email the address on the contact page. Markdown version for agents: /news/browser-use-1-0-cdp-python-repl-rewrite.md
Want these briefs in your Google Top Stories? Add Agentic Weekly as a preferred source on Google.
Newsletter
The Weekly Verdict. Free, every Monday.
A free email every Monday: the week's few big AI releases, then 5 short use cases for the new models, each with a prompt you can paste.
Free. Double opt-in. One click to unsubscribe.