Kimi K3 Launched: #1 on Frontend, #3 Globally
Kimi K3 tops Code Arena at 1679 points, surpassing Claude and GPT. Open-source with 2.8T parameters. Tested live—see how it performs.
“AI Disruption” Publication 10,200 Subscriptions 20% Discount Offer Link.
This morning, when I saw Arena AI update the Code Arena leaderboard, my first reaction was surprise.
The newly released Kimi K3 scored 1679 points, taking the global #1 spot and surpassing Claude Fable 5’s 1631 and GPT-5.6 Sol’s 1618.
Code Arena primarily evaluates front-end programming. The model must generate web pages based on user requirements, handling visual fidelity, interaction design, and functional implementation.
The leaderboard results come from millions of users participating in public blind tests. The same task is given to two anonymous models. Users experience both results and vote, and only after voting is completed are the model identities revealed.
Before the results are unveiled, users have no idea which company’s model they’re choosing. Brand influence is minimized, making the outcome far more credible. In other words, at least in front-end programming scenarios that emphasize visual presentation and interaction, K3 has taken the current top position.
A newly released open-source model effortlessly outperforming multiple closed-source flagships feels almost too good to be true. So we jumped on the hype and spent the night testing Kimi K3 to see if it’s the real deal.
Fun fact: The video below is not an official Kimi promotional clip. It’s an animation I generated by feeding the original trailer to K3 and having it recreate the entire thing from scratch.
From understanding the original video, reconstructing the web visuals, to handling camera work, motion effects, and interactions — the entire process was done by K3. Having the model create an intro video about itself is probably the most direct opening test.
