Vamsi Aribandi

training and shipping computer use agents @yutori_ai

[Vamsi Aribandi @varibandi]

@Jerr_Wu @HalluminateAI @oakhcft @ycombinator @orangecollectv @FTPartners @heavybit Congrats!!

[Lawrence Jang 🚣 COLM 26 @JangLawrenceK]

A bit late on this but congratulations to the @yutori_ai team for taking #1 on the MyPCBench leaderboard 🙂 Happy to see more work on pushing forward personalization + computer-use!

The evidence lies at https://t.co/4HZiGxR76l

[Rui Wang @theruiwang]

Navigator n2: 65.2% at $1.46.
Comparable performance costs 3x more. The premium costs 6.92x more.

[Vamsi Aribandi @varibandi]

Navigator N2 should be the default CUA model you try, but dont listen to me listen to the benchmarks! It also costs an order of magnitude less than Opus :)

https://t.co/8sBwBN69qV

[Devi Parikh @deviparikh]

Today we’re introducing Navigator n2. It’s a frontier computer-use model, with just 27B parameters.

[Vamsi Aribandi @varibandi]

@realYunfanYe Congrats, looks really cool!!

ROME

[Dhruv Batra @DhruvBatra_]

𝗡𝗮𝘃𝗶𝗴𝗮𝘁𝗼𝗿 𝗻𝟭.𝟱 “𝘀𝗼𝗹𝘃𝗲𝗱” 𝗢𝗻𝗹𝗶𝗻𝗲 𝗠𝗶𝗻𝗱𝟮𝗪𝗲𝗯: 𝟵𝟳.𝟯% 𝘀𝘂𝗰𝗰𝗲𝘀𝘀 𝗿𝗮𝘁𝗲.

While some teams self-report, this result is independently evaluated and verified by OSU NLP Group @osunlp and Careerflow Human Data Labs.

All benchmarks are transient attempts at measuring progress. Ultimately, what matters is how a model performs when people use it.

But there’s a sentiment online that computer-use models aren’t progressing quickly.

Not true.

In the last year, performance on Online Mind2Web has gone from ~40% success to basically saturated.

So what’s next?

Most computer-use/browser-use benchmarks are GUI-only. Models (including Navigator n1.5) now support hybrid actions — UI interactions (click, type, scroll) and programmatic actions (e.g., execute JS).

Ultimately, we’re headed to a world where computer-use models “agentify” the long-tail of the web.

https://t.co/E1B81aU0wa

[Vamsi Aribandi @varibandi]

@willcb Is it not clear that this debate is group based advantage estimation vs critic based? Anyone that’s reading more into it is either missing the forest for the trees or prioritizing the wrong experiments wrt algo exploration

[Vamsi Aribandi @varibandi]

@rohan_virani Lmfao

[Vamsi Aribandi @varibandi]

@Shahules786 @xdotli @adityak6798 @DavidAVenuto @GottliebEli @jackminong thanks for having me!

[Florian Brand @xeophon]

I don’t like Pulse in ChatGPT Pro. Its inaccurate, serves a lot of older information and is mostly slop. Which is weird, given how good ChatGPT is w/ search.

Scouts by Yutori (CUA startup) does those things better, imo.

[Vamsi Aribandi @varibandi]

@_arohan_ just confirming: this guy was at cal academy of sciences right?

[Vamsi Aribandi @varibandi]

https://t.co/ui0vWEti0G

[Vamsi Aribandi @varibandi]

co-driving RL for Navigator - our SOTA browser use agent - has been a professional highlight for me, have learned so much working on it with @CysuXiao. read the blog post (linked below) for more!

[Dhruv Batra @DhruvBatra_]

Introducing Yutori Navigator

31 years ago, the modern web era began with Netscape Navigator.

Today, we’re introducing Yutori Navigator — a web agent that autonomously navigates websites on its own cloud browser to complete tasks for you.

Navigator achieves pareto-domination over Gemini 2.5, Claude 4.5, and OpenAI Operator

• 10%-20% accuracy gains across benchmarks
• 2-3x faster
• Uniformly preferred in head-to-head human-evals

Simply put, the best web agent in the world

[Vamsi Aribandi @varibandi]

@alexgraveley congrats, looks cool!

[Vamsi Aribandi @varibandi]

@finbarrtimbers @natolambert another place this comes up is when it *always* uses .get on dicts instead of raw indexing. don’t think it learned that sometimes raising an error is good because it means something is wrong lol

[Abhishek Das @abhshkdz]

We're excited to launch Scouts — always-on AI agents that monitor the web for anything you care about.

[Vamsi Aribandi @varibandi]

@ParssaKyanzadeh looks really cool, can’t wait to try it!

[Vamsi Aribandi @varibandi]

@YiTayML agreed! (except blame being on you :))

we were definitely too grounded in the pre train-finetune and transfer learning paradigm, which was apparent when flan (and instructgpt) came out.

though personally it was a great learning experience for me so early in my career :)

[Vamsi Aribandi @varibandi]

@jxmnop bishop’s deep learning is recent and is very well written, imo should be the default textbook

https://t.co/nl7cbXBCpg

[Vamsi Aribandi @varibandi]

me in 2018 after training a cnn to classify dogs and cats with 60% accuracy

[Nick @nickcammarata]

honestly ai is so easy and neural networks are so simple. this was always going to happen to the first intelligent species to come to our planet. we’re about to learn something important about how universes tend to go I think, because I don’t believe we’re in a niche one