Moonshot AI Unveils Kimi K3.1 Pro, the First Pro Model in the Lineup, After the Kibitz Test

Moonshot AI has launched Kimi K3.1 Pro, the first model in its lineup to carry the Pro label, and in the same move revealed the identity of a mysterious Arena participant. For the past few days, an anonymous model with the codename Kibitz had been showing up in Arena's blind battles. Users coaxed SVG graphics, ready-made web pages and interactive 3D scenes out of it, and argued in the comments about whose work it was. The answer came overnight: Kibitz turned out to be Kimi K3.1 Pro.

How Kibitz Built an Audience Before Release

Blind testing in Arena is simple. A user submits a prompt, receives answers from two unknown models and votes for the better one. The names are revealed only after the vote. Labs have long used this channel to test fresh checkpoints without the weight of a brand. If a model wins when nobody recognizes it, that is a strong argument: people are praising the output, not the logo.

That is exactly what happened with Kibitz. Enthusiasts started "pulling cards", meaning they reran the same prompt until a good variant came up. They published the results: SVG illustrations, web pages and 3D scenes you can move around in and click on. Tasks like these take more than plausible-sounding text. The model has to keep in mind the document structure, coordinates and interaction logic, and still produce code that runs in a browser on the first try.

Why SVG and 3D Became the Main Argument

SVG has long been an unofficial test of spatial reasoning in language models. The image is described in code, so any error in paths, proportions or layer order is immediately visible. An interactive 3D scene adds another level: the model has to tie together geometry, lighting, controls and state. Models that handle such tasks confidently usually also do well in applied front-end coding, from interface prototypes to small games and visualizations.

That is how Moonshot AI staged its debut. Instead of a presentation with tables, it let the community find the model's strengths on its own, and then put its name on the result.

What We Know About the Release

• Developer: Moonshot AI (月之暗面), the creator of the Kimi lineup.
• Model: Kimi K3.1 Pro, the first in the lineup with the Pro label.
• Codename in testing: Kibitz.
• Testing venue: Arena's blind battles.
• Demonstrated use cases: SVG graphics, web pages, interactive 3D scenes.
• Moment of the reveal: the night of the release, after several days of anonymous appearances.

The appearance of the Pro label itself signals a shift in positioning. Until now, Kimi shipped as a single line of numbered versions. Now Moonshot AI is splitting it into tiers, as OpenAI, Google and Anthropic do, with a top model set apart from the mass-market one. For the Chinese market, where Moonshot AI, DeepSeek, Alibaba and ByteDance compete for both developers and enterprise contracts, this is a signal: Kimi has a flagship that can be sold separately.

How to Test the Model on Your Own Tasks

Flashy SVG and 3D demos do not guarantee that the model will hold up under your workload. So the sensible move is to run it on your own cases. For developers, that means generating an interface from a description, editing existing front-end code and porting markup to another framework. For designers and product teams, a test on icons, diagrams and infographics in SVG is useful. For game and educational projects, it is worth trying interactive scenes.

It is also worth comparing stability separately. Arena users regenerated answers many times, and the successful samples reached social feeds precisely because of that. If you need a working tool, look at how often the model hits the target on the first attempt, not just on the best of ten. That share will determine whether Kimi K3.1 Pro saves time in production or remains a pretty showcase.