Inside Kimi K3.1 Pro’s Three-Point Run Toward The Frontier

🔍 Read the full analysis: Inside Kimi K3.1 Pro’s Three-Point Run Toward The Frontier on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the little things that make your day delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Moonshot AI launched Kimi K3.1 Pro on October 23, and Artificial Analysis scored it 55.1 on its Intelligence Index, placing it fourth among the models in its October 24 comparison. The result puts Kimi three points behind the leading listed model, but Moonshot has not yet released the model weights or published a licence for them; the company says the weights are due by November 4.

Artificial Analysis lists K3.1 Pro as a mixture-of-experts model with 2.8 trillion total parameters and 104 billion active parameters. Its listing gives it a one-million-token context window and text and image input. The model became available through Kimi apps, Kimi Work, Kimi Code and the Kimi API.

In the benchmark table provided by Artificial Analysis, Kimi’s 55.1 score trails Claude Opus 5.5 at 57.6, Claude Sonnet 5.5 at 56.0 and Muse Wonder Full at 55.8. The result is higher than GPT-6 Astra at 52.7 and Gemini 4 Argon at 52.6. Artificial Analysis said Kimi would be the leading open-weights model once its weights are published; it is not yet available as a downloadable model.

Artificial Analysis lists Kimi’s API prices at $2 per million input tokens and $10 per million output tokens. It estimates about $2.11 per Intelligence Index task, compared with $5.98 for Opus 5.5 at its highest reasoning setting. Those figures are benchmark-specific estimates, not a guarantee of equivalent costs or results across real-world workloads.

A Smaller Gap, With Conditions

The score suggests that the measured performance gap between Kimi and leading proprietary models is narrow on this particular index. That could interest businesses weighing model quality against API costs, as well as developers who want the option to inspect or run model weights themselves. The cost estimate also points to a possible price advantage for some benchmark tasks, though it does not establish savings for every use case.

The distinction between an API launch and an open-weight release matters. Until the weights are published, users depend on Moonshot’s services and cannot independently run the model from its weights. Even after publication, the unpublished licence leaves open what users will be allowed to do. And a model of this scale would require substantial computing resources to run locally; availability of weights alone would not make self-hosting practical for most individuals or small organizations.

From Kimi K3 to K3.1

K3.1 Pro is described in the source as a more heavily post-trained version of Kimi K3. In Artificial Analysis’s comparison, Kimi K3 scored 43.6 on the Intelligence Index and 13% on Terminal-Bench 4.0. K3.1 Pro scored 55.1 and 67%, respectively. These are results from the named evaluations, not proof that it will outperform K3 on every task.

The reported results vary by category. Artificial Analysis measured K3.1 Pro at 1,846 Elo on GDPval-AA, behind Opus 5.5 at 1,866 and Muse Wonder Full at 1,857. On Humanity’s Last Exam, it scored 52%, compared with 47% for K3 and 61% for Opus 5.5. On Artificial Analysis’s AA-Omniscience evaluation, its measured hallucination rate was 44%, down from 53% for K3.

Moonshot has also promoted examples involving game development, Blender and CAD scripting, finance research, and a simulated walking project. The source identifies these as vendor demos or internal tests, rather than independently verified demonstrations. Moonshot’s reported 65% reduction in hallucinations comes from its own Kimi Hallu Bench Hard test; Artificial Analysis measured a smaller decline, from 53% to 44%, on its separate evaluation.

Weights, Licence and Real-World Results

The weights and their licence remain outstanding. Moonshot’s November 4 target is a promise, not confirmation that publication has occurred. The source does not say what licence will apply, whether release will include everything needed for deployment, or what hardware and operating costs would be for organizations seeking to run the model themselves.

Benchmark scores also do not settle how the model performs across everyday business tasks. Artificial Analysis’s results offer an external comparison, while Moonshot’s claims about complete game projects, CAD and finance rely in part on company-run tests and demonstrations. The source does not provide enough detail to independently assess those examples or establish that the reported hallucination rates predict performance in routine use.

The November 4 Weight Release

The next stated milestone is Moonshot’s promised release of K3.1 Pro’s weights by November 4. Publication would allow closer scrutiny of the model and make it possible to assess whether outside users can run it under the terms of its licence. Until the weights and licence are available, Kimi K3.1 Pro’s open-weight status remains prospective rather than established.

Further independent testing will help show whether its benchmark results carry over to specific coding, agentic and knowledge-work tasks. For now, readers can distinguish between the external scores reported by Artificial Analysis, Moonshot’s internally reported results, and the still-pending details of access and permitted use.

Key Questions

What is Kimi K3.1 Pro?

It is a model launched by Moonshot AI on October 23, 2026. Artificial Analysis lists it as a 2.8-trillion-parameter mixture-of-experts model with 104 billion active parameters, a one-million-token context window, and text and image input.

How did it rank against other models?

Artificial Analysis gave it 55.1 on the Intelligence Index, fourth among the models in the source’s comparison. It trailed Claude Opus 5.5, Claude Sonnet 5.5 and Muse Wonder Full.

Can users download Kimi K3.1 Pro now?

Not according to the source. The weights had not been published; Moonshot said it planned to release them by November 4. The source also says no K3.1 licence had been published.

Is Moonshot’s 65% hallucination reduction independently verified?

No. That figure comes from Moonshot’s internal Kimi Hallu Bench Hard test. Artificial Analysis measured a decline from 53% to 44% on its separate AA-Omniscience evaluation, a different result on a different test.

How much does Kimi K3.1 Pro cost to use?

Artificial Analysis lists API prices of $2 per million input tokens and $10 per million output tokens. Its estimated cost per Intelligence Index task is $2.11; actual costs depend on usage and the task.

Source: ThorstenMeyerAI.com