Browser Use (@browser_use) header
Browser Use (@browser_use) profile picture

Agents that use the Browser | 200k⭐️

San Francisco
Joined November 2024
Photo by Browser Use (@browser_use)Photo by Browser Use (@browser_use)Photo by Browser Use (@browser_use)Photo by Browser Use (@browser_use)Photo by Browser Use (@browser_use)Photo by Browser Use (@browser_use)Photo by Browser Use (@browser_use)Photo by Browser Use (@browser_use)Photo by Browser Use (@browser_use)Photo by Browser Use (@browser_use)
Browser Use retweeted
Replying to @gdb
@gdb codex CLI/app should use raw CDP (browser use CLI style), your harness inside codex CLI performs much worse right now
2
12
835
Browser Use retweeted
Browser Use 1.0.0 is basically 1 python REPL tool. > python + finish + finish_from_repl
Introducing Unharnessed, a new generation of harness. One tool: a Python REPL. Designed for the next generation of models. On Terminal-Bench 4.0, Unharnessed achieves a new state-of-the-art result.
5
3
1
74
4,626
GPT-6.2 Sol is the top model for browser use 👀 > 82%, up from 81% for GPT-6.1 Sol > 35% cheaper > 2x faster: 12 min instead of 22 min > 36% fewer steps
9
20
7
394
61,817
> why so fast: way fewer steps (median 67 vs 105 on GPT-6.1 Sol) and each step is quicker > fastest of the top models: nothing scores higher in less time > high reasoning, 180 shared tasks, same harness and judge for every model
1
17
2,738
the last eval made us realize our cache hit rate was kinda low (78%), so we fixed the harness now it's 96%. 33% cheaper ($0.67 -> $0.45 per task) and faster (5.0 -> 4.6 min), same score uv add "browser-use==1.0.4"
We measured 7 browser harnesses on 200 very hard tasks same model (gpt 6.1 terra medium), same tasks, same judge. only the harness changes > Browser Use Benchmark v2.1: 200 super hard long horizon tasks, mostly from our Odysseys v2 collab with CMU + a few extra on top > 6 browser tools inside codex (top harness on frontierharness): 62.4 to 73.5 > browser use 1.0 agent: 77.5
5
2
2
60
7,261
Introducing: @browser_use x @1Password Your agent can log in anywhere without ever seeing your password. > connect 1Password once > agent hits a login page > you approve it in 1Password > 1Password fills it in, agent keeps going Try it 🔗⬇️
9
11
4
160
9,017
We measured 7 browser harnesses on 200 very hard tasks same model (gpt 6.1 terra medium), same tasks, same judge. only the harness changes > Browser Use Benchmark v2.1: 200 super hard long horizon tasks, mostly from our Odysseys v2 collab with CMU + a few extra on top > 6 browser tools inside codex (top harness on frontierharness): 62.4 to 73.5 > browser use 1.0 agent: 77.5
13
5
6
150
17,240
Browser Use retweeted
Soon, you'll prefer our remote browsers over your local one. It'll feel like magic. 🪄
1
1
14
3,872
Bringing 996 Club vibe to Slovenia @browser_use
1
14
992
Browser Use retweeted
Browser Use cloud browsers can bypass anything. I will give you $100 amazon gift card if you find a website that blocks us 100% of time. comments ↓
176
13
12
560
100,242
Challenge?
Browser Use cloud browsers can bypass anything. I will give you $100 amazon gift card if you find a website that blocks us 100% of time. comments ↓
3
24
4,868
Browser Use is all you need
12
49
6,444
The entire Pareto Frontier is now dominated by Ember
Ran Muse Ember on hardest long horizon browser benchmark 👀 It's EXTREMELY cheap, pareto frontier. Tasks with 100s of browser steps cost CENTS, not USD > 63.9/100, same as GPT-6 Luna medium > 3 cents a task, 2.4x cheaper > best score under 5 cents congrats @alexandr_wang, cheap models that can actually drive a browser are huge
5
6
2
105
8,611
Try our cloud, we bypass everything
Replying to @browser_use
@browser_use has solved captchas. their stealth browser is so good that you essentially never get captcha blocked anymore. there are probably some very very lucrative ideas that stem from this.
6
5
257
15,190
Introducing: Browser Use 1.0.0: CDP is all you need 👀 We deleted 90% of the old library. Predefined state and action space make it worse. Now the agent just writes Python REPL on raw CDP. Simplicity wins. > same models, up to 50% better > 2-5x cheaper > 2-3x faster > SOTA web agent From now on we ship fast. Please pin with ==
50
85
20
1,175
202,099
If you need high concurrency for your agents browsers, let us know and we will help you. Our no.1 goal is to provide value.
1
3
837
Ran Muse Ember on hardest long horizon browser benchmark 👀 It's EXTREMELY cheap, pareto frontier. Tasks with 100s of browser steps cost CENTS, not USD > 63.9/100, same as GPT-6 Luna medium > 3 cents a task, 2.4x cheaper > best score under 5 cents congrats @alexandr_wang, cheap models that can actually drive a browser are huge
12
16
7
203
29,021
little browser agent hack: stop sending screenshots to find buttons. ask Chrome for the accessibility tree: cdp("Accessibility.getFullAXTree") > every element with role, name and node id > filter to links, buttons, inputs > DOM.getBoxModel -> click the center on Hacker News: 1,764 nodes -> 227 clickable, zero vision tokens Browser Harness agents work this way: github.com/browser-use/brows…
24
11
258
12,885