
Co-Founder & CEO @freesoloco | @ycombinator @zfellows | prev. @PennMandT
San Francisco, CA
Joined July 2022
- Tweets149
- Following332
- Followers1.1K
- Likes256
Eric Mao retweeted
Today, @JudgmentLabs opens for general access.
We believe long running agent systems will generate more exhaust data than the world can currently process.
Hidden within agent traces are rich records of how to make these systems better: what users ask for, when agents stumble, and how agents interact with the external world.
Judgment is the most performant monitoring and improvement layer for long-horizon agents.
We’re excited to see what you build!
Even though q3 ‘26 revenue isn’t explicitly stated, bottom of page 183 says it was ~86 times q4 ‘24 revenue which works out to ~16.9b.
For reference
q4 ‘25: 2b
q1 ‘26: 4.7b, +132%
q2 ‘26: 11.5b, +144%
q3 ‘26: 16.9b, +47%
Seems like the trends posted by openrouter and ramp were largely accurate. Interested to see if opus 5.5 is able reverse this trend. Even at 47% quarterly growth ant will still hit 100b arr by eoy.
Anthropic has filed an S-1 registration statement with the SEC related to its proposed IPO.
Read more: anthropic.com/news/s1-sec
Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.
1
271
Our old clado account tag was compromised and claimed by a third party after we rebranded to @freesoloco, now it seems to be shilling a coin that has nothing to do with us. Cc @Support
I’m surprised Jev for tabular data isn’t a bigger deal. Every ai spreadsheet company should be racing to rebuild their product ground up with Jev. When we were building people search last year we spent months optimizing the search algorithm. Here is Jev processing 100k rows for $2.50 in under 60s, no optimizations whatsoever. Public demo at spreadsheet.ericmao.com try it out with your own csv. A few more use cases I found interesting below
I borrowed a ton of concepts from the lotus paper. The idea was well ahead of its time. Credit to the core contributors: @lianapatel_ @sid_jha1 @pgasawa @melissapan Harshit Gupta
I was dealing with some insomnia over the weekend so I decided to teach astra + SO-101 arm (we'll call him BOTticelli) to draw. Through a series of real2sim2real training, calibration tricks, astra went from not being able to draw coherent shapes on Friday night to plausibly doodling better than me by Sunday.
There are many hardware problems with this setup (it was 5am) due to a over reliance on duct tape. This caused BOTticelli to draw like it has Parkinson’s. Interestingly Astra picked up this repeated flaw and recalibrated the simulation to simulated the shaky unsecured pen. In the 7 wonders drawing it actually learned in simulation more optimal stroke patterns containing the shakiness and transferred that to paper.
1
5
200
Nothing says resilient like getting deported for half a year and still giving it your all everyday. Having great cofounders is the hack that solves everything when shit hits the fan.
After getting put on administrative processing and being out of the US for more than 4 months, finally got my o1 visa!!!
So grateful to be building in the greatest country in the world with the greatest people in the world :)
1
9
907
Eric Mao retweeted
Today we're excited to open source Flash: a post-training stack for open weight models
Flash is designed to be cost-efficient and extendable. It handles training algorithms, compute provisioning, and model serving.
Use Flash via the CLI or self host with your own hardware.
Big deal for local models.
We introduce Maple-Preview, an open-source 20B-A1B ternary-weight reasoning LLM, SOTA in its weight class.
It solves IMO-level problems and runs at 200+ tokens/s on a Mac Mini M4, 5–16× faster than efficient models like Gemma 4, Qwen3.5, and gpt-oss.
deepgrove.ai/maple-preview
4
1,431
Its been just over a year since we were using 70b tokens a month running Clado's people search engine. Our product was bottlenecked by the breadth of inference we could do and the latency of the models (we were doing search in real time). Our solution was to post train a custom model to match the performance of the frontier on a fraction of the parameters. This project took our team of 3 a month of managing gpus, building training loops, and designing reward functions to get to the first usable model. Now you can watch tom build it from scratch in 30 minutes on 10 bucks.
Not every AI interaction is best served with a large frontier model.
There is a long tail of trillion-token use cases, from tagging to search, best served by a sub-10b-parameter model that runs in milliseconds and costs many orders of magnitude less than the frontier. We built Freesolo Flash to make training loops like SFT and RL easy and end-to-end completable through your coding agent.
4
12
1,941
Eric Mao retweeted
We’re launching @JudgmentLabs today and announcing $32M in funding.
As AI agents take on more of the work that creates economic value, they generate massive amounts of production data: the clearest record of how they behave with users, software, and the real world.
Judgment builds infrastructure for improving AI agents from production data.
Eric Mao retweeted
Chert (@Cherthq) is building the infrastructure for teams to send, receive, and automate conversations over iMessage using APIs.
They’re already working with B2C startups, customer experience teams, and sales teams to reach end users through trusted and conversational messaging.
Congrats on the launch, @GaryGao891219 & @ianyfong!
ycombinator.com/launches/QGN…











