Inside this story Context & sources

This article links to 3 sources. Source links are provided below so you can inspect the evidence yourself.

  • Moonshot’s new release is available through its hosted products; full weights are promised by 4 November.
  • Kimi Code access and context limits depend on the membership plan.
  • Our proposed test is an editable handover and a second revision; we have not run it.

The useful result is an editable project

Moonshot’s Kimi K3.1 Pro release brings a new coding and 3D-creation model to its apps, Kimi Work, Kimi Code and API. IT Home reported the launch on 24 October, Beijing time. The interesting question for a small creative team is what happens after the first attractive preview: can someone open the work, understand its structure and make the next change? [1] [3]

Moonshot says its 3D Make feature lets people preview and edit models online, download them for 3D printing, or export video. Those are different destinations for a result. A moving image may communicate an idea clearly, while an editable model is what lets the next person revise its shape or proportions. We have not tested those exports. [1]

Start with the file you need

Our suggested first brief would name the intended handover before describing the object. For a shop-display concept, for example, ask for a simple model with dimensions and clearly named components. Keep the reference and the prompt beside the output. Then ask another person to make one ordinary change. That exercise would tell a team more about the handover than a polished opening view.

The release also distinguishes using the hosted model from obtaining its full weights, the files needed to run the model on other infrastructure. Moonshot says the full weights will be released by 4 November. That remains a future commitment at this writing. It does not establish that the complete model can already be downloaded, or that a particular computer can run it. [1]

Check the account before the context window

Kimi Code’s current documentation lists K3.1 Pro and a Highspeed variant, with access and context allowances tied to membership plans. Its one-million-token context option is reserved for higher tiers. Context is the material a model can consider in a request; the headline capacity alone does not describe what a particular account receives. [2]

The documentation also recommends starting a new session when moving to the new model. We would keep that comparison tidy: use the same small brief, retain each result and record which model and account setting produced it. Moonshot’s speed and quality claims are useful questions for that exercise, rather than results we can reproduce from the launch page. [2]

Make the second edit part of the brief

For a South African studio or independent developer, our starting point would be an invented or already public project. Check the actual account access, export route and working terms before using client material. We have not verified South African billing, regional access or a local support arrangement, and an overseas launch cannot settle those details.

A prototype earns its place when it helps the next conversation move forward. The first render can make an idea easy to see; the retained project, references and a successful revision make it easier to continue. Kimi’s new release offers another route to try. The useful evidence would be the file and the second edit, not just the preview.

Sources & further reading

1. Moonshot: Introducing Kimi K3.1 Pro

Original announcement supplies availability, 3D Make and the promised 4 November weights release. Benchmarks and capability statements are vendor claims. No visible publication date; metadata dateModified is 24 October, which is not treated as the launch date.

2. Kimi Code: model configuration and plan boundaries

Undated documentation checked 25 October. Context allowance and model access depend on plan; vendor recommends a fresh session. Same company as the launch source, not independent corroboration.

3. IT Home: Kimi K3.1 Pro launch report — 24 October 2026

Qingyuan’s report is dated 24 October at 07:08 Beijing time and describes the launch early that morning. This establishes the dated announcement context, not an independent test of the model or its benchmark scores.

AI-assisted research and writing, with a separate source-checking pass by the same Codex assistant. Moonshot supplies product and capability claims; IT Home independently reports the announcement, not a measured model result. We have not used K3.1 Pro, exported a model or run the proposed revision trial. South African account access and billing have not been verified.

THE OPEN TABStay curious. Keep reading.