An Open-Weight Giant, A Near-Frontier Score: Kimi K3.1 Pro

🔍 Read the full analysis: An Open-Weight Giant, A Near-Frontier Score: Kimi K3.1 Pro on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get smart everyday buys 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, putting it three points behind the top-ranked model in the evaluator’s comparison. The release is being presented as a near-frontier model that may become a leading open-weight option, but its weights are not yet available and the company has not published a licence for K3.1.

Artificial Analysis lists K3.1 Pro as a mixture-of-experts model with 2.8 trillion total parameters and 104 billion active parameters. It accepts text and images and has a listed one-million-token context window. Moonshot made it available through the Kimi apps, Kimi Work, Kimi Code and its API. The evaluator’s October 24 figures put K3.1 Pro fourth, behind Claude Opus 5.5 at 57.6, Claude Sonnet 5.5 at 56.0 and Muse Wonder Full at 55.8.

The reported performance is not uniform across tasks. In Artificial Analysis’s Terminal-Bench 4.0 evaluation, which tests agentic coding work, K3.1 Pro scored 67%, compared with 13% for the earlier Kimi K3. On GDPval-AA, a simulated knowledge-work evaluation, it recorded 1,846 Elo, behind Opus 5.5 at 1,866 and Muse Wonder Full at 1,857. Its Humanity’s Last Exam score was 52%, while Opus 5.5 scored 61%.

Artificial Analysis lists API pricing at $2 per million input tokens and $10 per million output tokens. It estimates K3.1 Pro’s cost at $2.11 per Intelligence Index task, compared with $5.98 for Opus 5.5 at the models’ respective highest reasoning settings. These are evaluator estimates based on its testing and pricing inputs, not a guarantee of what a particular customer will pay. Moonshot’s separate claims about game development, CAD, financial tasks and hallucination reductions rely largely on company demonstrations or internal tests.

A Smaller Gap to Closed Models

If Moonshot releases usable weights under terms that permit meaningful deployment, K3.1 Pro’s score could narrow the performance gap between downloadable models and the leading proprietary systems. Artificial Analysis’s comparison places it three points behind Opus 5.5, a much closer result than the previous open-model score cited in the source material: Xiaomi’s MiMo-V2.6-Pro at 46.3. The comparison matters to organizations weighing model capability against control over where software and data run.

That possibility remains conditional. A benchmark result does not establish that an organization can run the model affordably, reproduce the score, or use it for commercial purposes. Artificial Analysis estimates a relatively low cost per benchmark task through the API, but self-hosting is a different calculation: a model with 2.8 trillion total parameters would require substantial infrastructure, and the applicable licence is not yet public. For users, the immediate change is API and app access; for developers seeking an open-weight model, the key development is still pending.

From Kimi K3 to K3.1 Pro

Kimi K3.1 Pro is described as a more heavily post-trained version of Kimi K3. In Artificial Analysis’s comparison, Kimi K3 scored 43.6 on the Intelligence Index, while K3.1 Pro scored 55.1. The largest reported difference is in Terminal-Bench 4.0, where the scores were 13% for K3 and 67% for K3.1 Pro. On Humanity’s Last Exam, the increase was smaller, from 47% to 52%.

Moonshot also promotes results from its own evaluations. It says K3.1 Pro reached 92.7% on BenchCAD with tools, compared with 96.2% for Claude Opus 5.5 and 95.9% for GPT-6 Astra in the company’s reported figures. Moonshot reports an internal claim-level hallucination rate of 8.3% on Kimi Hallu Bench Hard, down from 23.4% for K3. The company says that test is deliberately designed to provoke errors and that its results do not represent error rates in ordinary use. Artificial Analysis measured a decrease on its separate AA-Omniscience test, from 53% for K3 to 44% for K3.1 Pro—a different measure from Moonshot’s internal figure.

“The Kimi Hallu Bench Hard test is deliberately crafted to provoke errors, and the rates do not represent overall error rates in daily use.”

— Moonshot AI

Weights, Licence and Real-World Results

The weights have not been released; Moonshot says they are expected by November 4. The source material does not establish whether that date will be met, what licence will apply, or whether the terms will allow commercial use and modification. Until those details are public, describing K3.1 Pro as available open weights would be premature.

Benchmark scores also leave practical questions unanswered. The figures do not show how the model will perform across all production workloads, how results vary with different prompts and tools, or what it would cost to operate at scale. Moonshot’s claims about building complete games, creating parametric models and improving financial work are supported primarily by company demonstrations and internal evaluations. The source material does not provide independent verification of those demonstrations. Artificial Analysis’s independently run results offer a separate comparison, but they do not validate every vendor claim.

The November Weight Release

The next milestone is Moonshot’s promised November 4 weight release. Developers and businesses will be able to assess the actual files, licence and deployment requirements if the company publishes them as planned. Those details will determine whether the model can be downloaded and adapted under usable terms, rather than accessed only through Moonshot’s services.

After release, independent testing and reports from organizations running the model locally could clarify its hardware needs, performance and costs outside the API. Until then, the confirmed picture is narrower: K3.1 Pro is available through Moonshot’s products and API, Artificial Analysis has reported strong benchmark results, and open-weight availability remains a promise rather than a current fact.

Key Questions

What is Kimi K3.1 Pro?

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

How did K3.1 Pro score against leading models?

Artificial Analysis gave it 55.1 on its Intelligence Index, ranking it fourth in the reported comparison. It trailed Claude Opus 5.5 at 57.6, Claude Sonnet 5.5 at 56.0 and Muse Wonder Full at 55.8.

Can users download the weights now?

No. Moonshot has promised the weights by November 4, but they had not been released in the source material’s October 24 update. The company had also not published a K3.1 licence.

What does Kimi K3.1 Pro cost through the API?

Artificial Analysis lists a price of $2 per million input tokens and $10 per million output tokens. Its estimated $2.11 cost per Intelligence Index task is a benchmark-based comparison, not a prediction of every customer’s costs.

Are Moonshot’s claims about hallucinations independently confirmed?

Moonshot reports an internal reduction from 23.4% to 8.3% on its Kimi Hallu Bench Hard test and says that test is designed to provoke errors. Artificial Analysis measured a separate change on AA-Omniscience, from 53% for K3 to 44% for K3.1 Pro; the two figures use different evaluations.

Source: ThorstenMeyerAI.com