پن کیا گیا ٹویٹ
Hi all, I am a 3rd year undergrad at Stanford studying computational physics. I also lead agent evaluations at Browser Use.
I'm starting this X account as a place to voice my thoughts on AI and agentic developments
English
Alexander Yue
192 posts
Alexander Yue
@Alezander907
Physics & CS @ Stanford SLAC | Agents @ Browser Use
Stanford شامل ہوئے Temmuz 2017
54 فالونگ563 فالوورز
@Alezander907 how do the open-weight ones hold up on multi-step authenticated flows? running agents on ~300 supplier portals (login + 2FA) and trying to figure out if the cheaper models break too often to be worth it at scale
English
Anthropic has now caught on with slack agents
Alexander Yue@Alezander907
I have built custom agents in our browser-use slack that have boosted my productivity 3x Each one has persistent learning, builds tools, schedules self, edits own source code. Different permissions for each They can mention and call each other. The overseer manages all
English
@sjsjsjsjsjko @larsencc Make and account and dm me what email you use and I will give you credits
English
Hear me out... Browser Harness but in the Cloud (beta).
Built on:
> Browsercode (thanks @Alezander907)
> AWS AgentCore
> Custom Control Plane
Try it in the UI, or comment API V4 for early API access ↓
English
@LastResort48 @larsencc Make an account and dm me what email you use I’ll give you credits
English
@CastelMaker @larsencc Make an account and dm me what email you use I’ll give you credits
English
@martinbowling @larsencc Make an account and dm me what email you use I’ll give you credits
English
English
Browser Harness in the cloud is live (beta).
Break it so @larsencc and @Alezander907 can fix it 🫶
Larsen Cundric@larsencc
Hear me out... Browser Harness but in the Cloud (beta). Built on: > Browsercode (thanks @Alezander907) > AWS AgentCore > Custom Control Plane Try it in the UI, or comment API V4 for early API access ↓
English
Human memory is still the best memory for agents.
Seek to understand everything in your company. Give your agents the context they need.
It wouldn’t be better to replace this human layer with LLM memory. Maybe a faster search tool would help though
English
@_halshin Opus 4.8 gives more refusals about doing browser tasks
English
@Alezander907 Great to see this beating GPT-5.5, but why is the benchmark against Opus 4.7 and not Opus 4.8?
English
GLM 5.2 is a huge improvement for browser agents, offering near opus level score, beating GPT 5.5
Minimax M3 is a sonnet level score at just $0.30 input, my new best value model (cheaper than deepseek v4 pro)
Kimi k2.7 is a +9% improvement from k2.6 but is outclassed by M3
English
@watchereth_ Browser Harness is a huge new way to use browser - but just tools, no agent
BrowserCode is a opencode fork with browser harness included
Browse use v4 is running BrowserCode on our cloud for you, no installs required, looks like a chat app
English
Try it now, completely open source: github.com/browser-use/br…
Russ Salakhutdinov@rsalakhu
Congrats to the @browser_use team for taking the #1 spot on Odysseys, a highly challenging benchmark for long-horizon web agents: odysseys-website.pages.dev/leaderboard Odysseys evaluates realistic, multi-hour web workflows that require sustained planning, memory, reasoning, and verification across numerous websites and tools, far beyond short single-step browser tasks. Exciting progress toward truly capable long-horizon agents.
English
Alexander Yue ری ٹویٹ کیا
Congrats to the @browser_use team for taking the #1 spot on Odysseys, a highly challenging benchmark for long-horizon web agents:
odysseys-website.pages.dev/leaderboard
Odysseys evaluates realistic, multi-hour web workflows that require sustained planning, memory, reasoning, and verification across numerous websites and tools, far beyond short single-step browser tasks.
Exciting progress toward truly capable long-horizon agents.
English
Try v4 in Browser Use Cloud now!
Highest scoring agent of all time, now fully hosted
Browser Use@browser_use
Our new agent can find exactly where you are on a map 🤯 We asked Browser Use v4 to play GeoGuessr... It guessed the exact location within 50 km by > Analyzing the 3D view in Google Maps > Finding clues in the environment We just released it. Try it for yourself ↓
English
BrowserCode has now been verified at the top of the Odysseys leaderboard. And its the same capability in the new browser-use version we released!
English
