Kimi K3.1 Pro Posts A Frontier-Near Result As Release Wait Continues

🔍 Read the full analysis: Kimi K3.1 Pro Posts A Frontier-Near Result As Release Wait Continues on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Moonshot AI’s Kimi K3.1 Pro scored 55.1 on Artificial Analysis’s Intelligence Index, ranking fourth in the evaluator’s October 24 comparison and three points behind the leading score. The model went live in Moonshot’s apps and API on October 23, but its weights are only promised by November 4, and the company has not published a licence for them.

Artificial Analysis lists K3.1 Pro as a mixture-of-experts model with 2.8 trillion total parameters and 104 billion active parameters. It supports text and image input and has a 1-million-token context window. Moonshot describes it as its most capable “3T-class open-weight model,” but until the weights are released, users cannot independently download and run them.

In the evaluator’s comparison, K3.1 Pro’s 55.1 score put it behind Claude Opus 5.5 at 57.6, Claude Sonnet 5.5 at 56.0 and Muse Wonder Full at 55.8. Artificial Analysis said K3.1 Pro would lead open-weight models once its weights are published. That is a conditional assessment, not a ranking of a model already available to run from released weights.

Artificial Analysis listed API prices of $2 per million input tokens and $10 per million output tokens, with an estimated $2.11 per Intelligence Index task. Its listed estimate for Opus 5.5 was $5.98 per task at the evaluator’s chosen settings. Those figures reflect the benchmark’s pricing and task assumptions, not a universal cost for every workload.

A Smaller Gap to Closed Models

The score puts K3.1 Pro close to several leading proprietary models in this particular evaluation. If Moonshot publishes usable weights under a licence that permits the expected forms of use, developers and organisations could gain more control over deployment, including the option of running a model on their own infrastructure. That could matter for teams with data, latency or customisation requirements that are difficult to meet through a hosted API.

The caveats are material. The weights are not yet released, and an unpublished licence leaves unresolved what users will be permitted to do. Artificial Analysis also lists 2.8 trillion total parameters, making self-hosting a substantial infrastructure undertaking; “open weight” would not mean that running the model is inexpensive or straightforward. The benchmark is evidence about measured performance, not proof that K3.1 Pro will match closed models across every task or production setting.

The results also point to uneven gains. Artificial Analysis measured a 67% Terminal-Bench 4.0 score for K3.1 Pro, compared with 13% for K3, while its Humanity’s Last Exam result rose from 47% to 52%. Those measures suggest a particularly large improvement in the tested agentic coding tasks, but smaller progress on that knowledge benchmark. Performance should be read by task and test, rather than reduced to one headline score.

What the Benchmarks Compare

Kimi K3.1 Pro is described in the source material as a more heavily post-trained version of Kimi K3. Artificial Analysis’s October 24 model-page figures compare models at their highest listed reasoning settings. In that comparison, K3.1 Pro’s 55.1 was above GPT-6 Astra at 52.7 and Gemini 4 Argon at 52.6, and above the earlier Kimi K3 score of 43.6.

Moonshot has also presented demonstrations involving game engines, Blender and CAD work, finance tasks and reduced hallucinations. These are company-reported examples and results, not all independently verified demonstrations. For example, Moonshot reported that a test of its own Kimi Hallu Bench Hard showed claim-level hallucinations falling from 23.4% for K3 to 8.3% for K3.1 Pro, a reduction of about 65% on that test.

Artificial Analysis measured a decline in its separate AA-Omniscience evaluation, from 53% to 44%. The two figures come from different tests and should not be treated as interchangeable. Moonshot itself cautioned that its hallucination test was deliberately designed to provoke errors and that its rates do not represent overall error rates in everyday use.

“Kimi K3.1 Pro is its most capable “3T-class open-weight model.””

— Moonshot AI

Weights, Licence and Real-World Results

The source material says Moonshot promised the weights “by Nov 4”, but does not establish whether they have since been released. It also does not provide a K3.1 Pro licence. Until those details are available, the model’s practical status as a downloadable, reusable open-weight system remains unresolved.

Benchmark scores do not establish how the model performs across every real-world use case, and the information provided does not include independent verification of Moonshot’s game, CAD, finance or internal hallucination demonstrations. The reported 67% Terminal-Bench score and 55.1 Intelligence Index result are measurements from Artificial Analysis; they do not guarantee comparable outcomes in a company’s own tools or workflows.

It is also unclear what hardware and operating costs would be required to run the full model locally. Its total parameter count is 2.8 trillion, but the source does not give a deployment configuration or estimate of the resources needed. Hosted API prices are available, but those do not answer the separate question of self-hosting costs.

The November 4 Weight Release

The next stated milestone is Moonshot’s promised release of K3.1 Pro’s weights by November 4. Readers and developers will then be able to check whether the files are published as promised, review the accompanying licence and assess what forms of use it allows. The source material does not identify a separate publication date for the licence.

After release, independent testing of the downloadable model could clarify whether its benchmark standing carries over when run outside Moonshot’s hosted services. Until then, the clearest confirmed picture is a model already available through Moonshot products, with a strong Artificial Analysis score and open-weight availability still pending.

Key Questions

What is the main news about Kimi K3.1 Pro?

Artificial Analysis scored Kimi K3.1 Pro at 55.1 on its Intelligence Index, placing it fourth in its October 24 comparison. The model went live in Moonshot’s apps and API on October 23, while its weights remain promised for a later release.

Can users download the model weights now?

Not according to the source material. Moonshot said the weights would be released by November 4; the material does not confirm that they have been published.

What licence will apply to the weights?

No K3.1 Pro licence had been published in the source material. That leaves the permitted uses and redistribution terms unclear.

How close is the model to the top-ranked result?

Its Artificial Analysis Intelligence Index score was 55.1, compared with 57.6 for Claude Opus 5.5 in the same comparison. The scores describe that evaluator’s test, not every model capability or workload.

Are the reported hallucination gains independently verified?

Moonshot’s reported reduction from 23.4% to 8.3% came from its own Kimi Hallu Bench Hard test. Artificial Analysis measured a different result on its AA-Omniscience evaluation, where the rate fell from 53% to 44%. The figures use different tests and are not directly comparable.

Source: ThorstenMeyerAI.com