Meet Instinct — our open-weight System One models built for decisions, not generation.
87.45% accuracy.
5.5× faster in our test.
Up to 76% cheaper.
No prose. No JSON. Just calibrated probabilities agents can act on.
Built for the shift from AI that talks to AI that decides.
JEV put System One models on the map.
We think the category should be open.
So we built Instinct.
87.45% accuracy vs Jev's 86.58%.
5.5x faster in our test.
Up to 76% cheaper.
All weights open.
And it doesn't generate a single word.
The demo below? Generated by our AI agent ↓
AI Commerce Hackathon · Oct 3 · San Francisco
Bring the idea you've been sitting on. Build with mentors, demo to investors, and compete for a spot on the main stage.
ZooWork × @aivalleyio
The Walt Disney Family Museum
Apply to join: luma.com/6bbloggr
ZooClaw is now ZooWork.
What’s changed: our focus has expanded to helping experts deliver AI agents and businesses run AI on an open stack they control.
What hasn’t: AI should make life better.
Seedance 2.5 is now live on ZooClaw. 🎬
Now you can generate videos up to 30 seconds, feed in up to 30 images, 10 videos & 10 audio clips, or even start with audio alone.
And with MOV export, your generations can go straight into the editing workflow.
If people keep asking, “Can I pick your brain?”—there may be a business hiding in the answer.
See what it could become with ZooClaw.
ZooData finished #1 on Product Hunt. 🥇
Thank you to everyone who tested it, challenged the idea, and told us what their agents actually need.
Now we get back to building.
1,000 free credits are still live: zoodata.ai
🤖 Made with AI
ZooData is #1 on Product Hunt right now!
A lot of builders share the same frustration: agents are still chewing through data made for people.
If that’s your problem too, come see what people are testing: producthunt.com/products/zoo…
AI agents don’t need more pages to read. They need better data to reason with.
@ZooDataAI turns any URL into agent-ready JSON — clean extraction, fewer tokens, smarter workflows.
Start with 1,000 free credits (no card required) 👉🏻 zoodata.ai
We got tired of watching agents chew through messy web pages.
Meet ZooData tomorrow.
Clean data for agents. Less mess to parse.
When the benchmark gets benchmarked.
GPT Deep Research is already top-tier. But in this side-by-side test, ZooClaw scored higher.
Topic: the first job AI agents should actually replace in 2026
ZooClaw: 87.5/100
GPT: 82.5/100
Cleaner is not always deeper.
86M params. Every benchmark, in-domain and OOD — beats specialized baselines and a 1B backbone.
Full FT + distillation + WiSE-FT. No tricks.
arxiv.org/abs/2606.27708
ZooClaw has always been building vertical skills for agents.
Fashion is one of the hardest domains to get right.
So we built the strongest one.
ZooClaw-FashionSigLIP2 is now open-source: vertical SOTA for fashion product understanding.
huggingface.co/srpone/zoocla…
Seedance 2.0 mini is now supported in ZooClaw:
fast video generation, built for high-volume content, with ~50% lower token cost.
You watch the game.
ZooClaw ships the videos.
World Cup season productivity hack:
Open match
Pretend to work
Let ZooClaw handle the actual work
#WorldCupPredictions
🤖 Made with AI
Fable 5 is powerful.
But inside ZooClaw, it gets a full research team.
ZooClaw’s built-in skills inspect the reference site, understand the design system, structure, and interaction details, then Fable 5 recreates the webpage in one click.
Now live in ZooClaw.
ZooClaw has always been building vertical skills for agents.
Fashion is one of the hardest domains to get right.
So we built the strongest one.
ZooClaw-FashionSigLIP2 is now open-source: vertical SOTA for fashion product understanding.
huggingface.co/srpone/zoocla…
Stop overpaying for AI video.
ZooClaw now supports Seedance 2.0 mini: ~50% lower cost than Seedance 2.0 for high-volume ecommerce, UGC, marketing, and effects videos.
More videos. Less spend.
















