MBZUAI's K2 Horizon 375B Leaps 30 Points and Ships Its Weights Free
As seen on the 24/7 Wall St. homepage on September 3, 2026.
A 30 point benchmark jump from a UAE lab giving its weights away puts more pricing pressure on closed US frontier models, and every one of these free releases still needs rented compute to train and serve.
K2 Horizon 375B A23B, a new open weights model from UAE's MBZUAI, scores 47 on the Artificial Analysis Intelligence Index, with relatively strong agentic performance and a 30 point jump over its predecessor K2 Horizon 375B A23B is an open weights Mixture-of-Experts model with https://t.co/um82zkFyIs
- Replies2
- Reposts8
- Likes71
Continue ReadingShow less
K2 Horizon 375B A23B is a Mixture-of-Experts model released with open weights by MBZUAI, the Abu Dhabi-based AI research institute, and it scores 47 on the Artificial Analysis Intelligence Index, a 30-point improvement over its predecessor.
The 30-point jump signals that a non-US lab is closing the capability gap at a pace that matters, and releasing the weights freely removes the pricing floor that closed frontier models depend on.
Sponsored
Twelve Tabs, One Thesis
Your Research Resets Every Morning
The quote page in one tab. Filings in another. A chart you rebuilt from scratch, a transcript you never went back and found, a screener whose settings you will redo next week. Nothing you built yesterday is still there.
AlphaSpace replaces all of it with one screen you arrange yourself. Earnings calendar, estimate versus actual, the call transcript, live news, your own charts, every panel wired to whatever ticker you click. Close the browser and it is all still sitting there tomorrow.
See What a Built View Looks Like →
(Sponsor)
The model's relatively strong agentic performance is a particular pressure point, because agentic workloads are where enterprises are actively choosing which models to pay for right now.
Every open-weights release like this still requires significant compute to train and serve, so the cost pressure lands on cloud providers and API vendors more than on the labs giving weights away. The question for closed-model incumbents is how long benchmark leads hold when a credible open alternative keeps arriving for free.