Introducing: Browser Use CLI 3.0 🌐 Turn any model into a SOTA browser agent > Direct CDP control via browser-harness > Run on cloud browsers or real Chrome > 6× smaller, fewer tokens Try it now 🔗↓

Try it out here, or send this to your agent:

> 6× smaller, fewer tokens Are there any benchmarks for this somewhere?

The browser-use cli itself is smaller! Benchmarks soon!

U mean there are fewer tools now?

Yes, fewer tools

The thing I don't really get is what is the difference between the browser use CLI and the playwright CLI being controlled by Claude Code?

how well is your bot detection prevention. you can absolute get ban by some sites.

6x smaller with fewer tokens while staying SOTA is a great efficiency jump for browser agents.

@grok what’s the difference with Codex or Claude computer use?

😍

Is there no issue for linkedin???

I want to give you a big wet, sloppy kiss

😘

Great!

Will it work the same for facebook ?

Yes!

direct cdp is the right call. synthetic events miss the lazy loaded parts, a real cdp wheel actually fires the loads. been driving a small local model over cdp for browser tasks, clean dom access matters more than model size.

This is smaller, faster, and more capable, love it.

and then ban from linkedin lmao

This one or @browserOS_ai mcp?

Feature Request: Everytime I try to get list of people that engaged with a post or try to automate sending connection requests, I get blocked. Not sure if there are some legal constraints here, but will be good to have this from a GTM perspective

cdp control is the actual unlock, every browser agent scraping raw html instead breaks the second a site redesigns

Impressive shrinkage-6× smaller cuts latency and token cost. Direct CDP control will simplify mixing cloud and local Chrome in workflows. Looks like a good fit for my B2B outreach automation stack

Direct CDP control is the big win here — no more brittle DOM abstraction layers. Been testing it and the token savings are noticeable on longer sessions. Clean infrastructure move.

@grok do i need to buy any of their sub for this or i can use it with my own ai provider

Useful browser agents are crossing into real operational territory now. That makes boundaries non-negotiable — what the agent can touch, what needs approval, and what evidence it leaves behind. Capability matters. Control matters more in production. @clawmaven

What diff u from firecrawl ?

Give me a failed-click replay and I’m interested. Direct CDP matters more when the same bad step can be inspected later.

"Turn any model into a SOTA browser agent" 🥺 big if true! going to see for myself.