Moonshot’s launch post for Kimi K3.1 Pro (23 October 2026) claims strong finance gains and a 3D Make feature, but it trails Claude Opus 5.5 on CAD. It also cut cyber capability on purpose for the open-weight release.
Every number in this guide is Moonshot’s own claim unless it says otherwise, and none of it tests Arabic or your own tasks. For the independent score, prices and the weights promise, see our Kimi K3.1 Pro analysis: Artificial Analysis ranks it fourth on its Intelligence Index with 55 against 58 for Claude Opus 5.5.
The short answer
- 3D Make: live on kimi.ai and the Kimi app. You preview and edit 3D models online, download them for 3D printing, or export videos and Live Photos.
- CAD (BenchCAD, with tools): K3.1 Pro scores 92.7 against 96.2 for Claude Opus 5.5 and 95.9 for GPT-6 Astra, so it trails both.
- Finance: OfficeQA Pro v2 64.4 (described as state of the art); Frontier Finance 58.6.
- Hallucination (internal stress test): the claim-level rate fell from 23.4% (K3) to 8.3%.
- Cyber: cyber data was removed for the open-weight release. On the internal xdev-bench-hard it scores 81.0%, against 90.5% for the checkpoint without the measures.
- Weights: promised “By Nov 4, 2026” and not published at the time of reading.
What does 3D Make do, and how good are the CAD numbers?
Moonshot says K3.1 Pro “particularly excels at parametric modeling”. That means models defined by parameters, such as mechanical parts, architectural models, configurable products and custom furniture.
Without MCP connectors it can write and run scripts for Blender and CAD software. Moonshot says it also works with Unity, Unreal Engine and Godot, together with web search and 3D generation. For what MCP connectors are, see What is MCP.
BenchCAD is a 1,000-file subset scored by voxel IoU. With tools, K3.1 Pro scores 92.7, behind Opus 5.5 (96.2) and GPT-6 Astra (95.9). By our subtraction the gap to Opus 5.5 is 3.5 points. Without tools, the chart shows K3.1 Pro at 72.8 and Opus 5.5 at 73.0, essentially level. GPT-6 Astra has no score there. The no-tools numbers appear in the chart only.
The meaning: tools lift K3.1 Pro by about 20 points (92.7 minus 72.8, our subtraction), so judge it as an agent with tools, not as a bare model.
How strong are the finance claims?
| Measure | K3.1 Pro | K3 | Source in the post |
|---|---|---|---|
| OfficeQA Pro v2 | 64.4 | not given | Body text |
| Frontier Finance | 58.6 | not given | Chart |
| Financial Modeling (internal) | 57.6 | 50.4 | Gain of 7.2 points in text; values in chart |
| Financial Research (internal) | 70.4 | 58.8 | Gain of 11.6 points in text; values in chart |
| Business Operation (internal) | 37 | 8.9 | About 4.2 times K3 in text; values in chart |
Moonshot built the three internal benchmarks. They show improvement over its own previous model, not a comparison with rivals. The biggest jump is Business Operation, from 8.9 to 37, but it starts from a very low base. Do not read these tables as advice about financial decisions.
Is the hallucination drop a daily error rate?
No. On Kimi Hallu Bench Hard, Moonshot says the claim-level hallucination rate fell from 23.4% for K3 to 8.3% for K3.1 Pro. The post itself notes that this is a stress test and that the rates do not reflect typical daily error rates. It gives no figure for how often you will see a wrong answer in normal use. Keep checking facts that matter.
What is the cyber trade-off?
For the open-weight release, Moonshot says it identified and removed cyber-related data from both mid-training and the reinforcement-learning environments, while keeping general-purpose coding tasks.
On its internal xdev-bench-hard (pass@3), the checkpoint without these risk-reduction measures solved 90.5% and the final K3.1 Pro reaches 81.0%. Moonshot states that the model’s cyber capability “remains behind” that of frontier proprietary models that have been publicly available for at least three months. That is the company’s assessment of its own model, not an independent security test.
Where can you use it, and what about Terminal-Bench?
Moonshot reports 65.7% on Terminal-Bench 4.0 against 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5, run with the Claude Code harness. Artificial Analysis reports 67% under a different setup, so keep the two numbers apart.
The model is available in Kimi Work (desktop app 3.1.0 or later for Windows and Apple silicon Macs), in Kimi Code through the /model command, and through the API. Kimi Enterprise administrators must enable it because access is off by default. Price comparisons are in Cheaper Tokens, Bigger Bills.
Who is this for, and what should you do next?
It suits teams that build 3D or CAD workflows and want an open-weight option. If you need the best CAD score today, Moonshot’s own table points to Opus 5.5 or GPT-6 Astra.
We suggest treating the post as a feature list and testing the model on a few of your own tasks, as in how to run a one-afternoon AI model evaluation. The numbers above measure CAD, finance and cyber tasks Moonshot chose. None of them measures writing quality in Arabic or in your domain.
Questions and answers about the new Kimi K3.1 Pro claims
Is Kimi K3.1 Pro the best model?
No claim of that kind is supported here. Artificial Analysis ranks it fourth (55 against 58 for Claude Opus 5.5), and on CAD it trails Opus 5.5 and GPT-6 Astra in Moonshot’s own table.
What can 3D Make export?
You can preview and edit 3D models online, download them for 3D printing, or export videos and Live Photos.
Why did Moonshot cut cyber data?
The post says it removed cyber-related data for the open-weight release. It reports lower cyber scores as a result: 81.0% against 90.5% on its internal test.
Can my team switch it on in Kimi Enterprise?
Only an administrator can. Access is off by default, so ask your admin to enable the model.
Terms in plain language
- Open weights: the trained model files are published so others can download them.
- Hallucination: a claim the model states that is not true.
- Parametric modeling: building a shape from adjustable parameters such as length or width.
- Benchmark: a fixed set of tasks used to compare models.
Sources
- Moonshot AI: Kimi K3.1 Pro (read 26 October 2026)
- Artificial Analysis: Kimi K3.1 Pro launch benchmarks (23 October 2026)
Facts read on 26 October 2026.
