Kimi’s new model wants to build the scene, not just show you a picture

From a picture to a project

Moonshot AI launched Kimi K3.1 Pro on 23 October. Its launch report puts 3D creation alongside coding and office work: the new 3D Make experience on Kimi’s website lets people preview and edit models, then download a model for 3D printing or export a video.

The system can write and run instructions for tools such as Blender, which builds 3D scenes, and CAD software used for precise models. Moonshot’s examples show what it wants that workflow to become. They are demonstrations, rather than evidence that a generated shop, object or installation is ready to manufacture.

That distinction matters for creative teams. A convincing render answers “does this look interesting?” An editable project opens another question: can a designer change its proportions, materials and layout without starting again?

A strong test result, with a narrower meaning

Artificial Analysis independently gives K3.1 Pro a score of 55 on its Intelligence Index. That places it near the leading models in this snapshot. The index combines ten evaluations across areas including reasoning, coding and work tasks; it does not measure whether a brand’s 3D campaign scene is accurate or easy to revise.

Treat the result as a reason to investigate, rather than a substitute for a real brief. Before using a project in a client presentation, inspect the geometry, check the requested details and make an actual revision. For physical production, a specialist still needs to check dimensions and suitability.

Available to use; not yet available to download

Hosted access is live through Kimi’s products and API. Moonshot promises downloadable weights—the files needed to run the model yourself—by 4 November. Teams that require their own infrastructure therefore have a different start date from those using the hosted service.

Kimi Code’s release notes put Pro access on Plus/Moderato plans and above, with its largest context allowance and Highspeed option on higher tiers. API usage is billed separately from subscriptions; the quick-start guide requires at least a $1 top-up to unlock flagship API access.

The published API rates are $2 for a million input tokens and $10 for a million output tokens, excluding tax. Tokens are the chunks of text a model reads and produces. Highspeed doubles those rates; cache-related charges have their own conditions in the full pricing table. A cheap first draft and an economical finished project are still different things.