Kimi K3.1 Pro: A Giant Model With A Near-Frontier Score

🔍 Read the full analysis: Kimi K3.1 Pro: A Giant Model With A Near-Frontier Score on ThorstenMeyerAI.com

Age 18–24?Offer from Amazon

Prime made for students and young adults

• Fast, free delivery for dorm and study essentials
• Prime Video and Amazon Music included
• Member-only deals

Try Prime for Young Adults →Free trial for eligible 18–24 year olds

As an affiliate, we earn on qualifying purchases.Hide

Moonshot AI launched Kimi K3.1 Pro on October 23, and an evaluation by Artificial Analysis ranked it fourth on its Intelligence Index, with a score of 55.1. That puts it three points behind the highest score in the listed comparison, but its promised weights are not yet available and the model’s licence has not been published.

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 a 1-million-token context window and text and image input. Moonshot made it available through the Kimi apps, Kimi Work, Kimi Code and its API. The company describes it as a more heavily post-trained version of Kimi K3.

On Artificial Analysis’s Intelligence Index, K3.1 Pro scored 55.1, behind Claude Opus 5.5 at 57.6, Claude Sonnet 5.5 at 56.0 and Muse Wonder Full at 55.8. The same evaluation put K3.1 Pro’s estimated cost at $2.11 per index task, compared with $5.98 for Opus 5.5 at its highest reasoning setting. API prices listed for K3.1 Pro are $2 per million input tokens and $10 per million output tokens.

Artificial Analysis measured a marked increase on its Terminal-Bench 4.0 test: K3.1 Pro scored 67%, compared with 13% for Kimi K3. On the GDPval-AA knowledge-work evaluation, it scored 1,846 Elo, behind Opus 5.5 at 1,866 and Muse Wonder Full at 1,857. Its score on Humanity’s Last Exam was 52%, while Opus 5.5 scored 61%.

A Smaller Gap to Closed Models

The score places K3.1 Pro close to several leading proprietary models in this particular evaluation, while its listed cost per index task is lower than Opus 5.5’s. For businesses comparing model performance and API spending, that combination makes the launch relevant, especially for agentic coding and knowledge-work tasks. Index results do not establish that one model will be better for every workload; performance can vary by task and evaluation method.

The bigger potential shift depends on access to the weights. Artificial Analysis described K3.1 Pro as set to become the leading open-weight model once its weights are published. If released under terms that permit the uses organisations need, weights could give some developers more control over deployment and data handling. But a 2.8-trillion-parameter model is a substantial infrastructure undertaking, and the licence and practical hardware requirements will shape whether that possibility is useful outside large operators.

The Benchmark Gap and Release Conditions

Before K3.1 Pro’s launch, Artificial Analysis listed Xiaomi’s MiMo-V2.6-Pro at 46.3 as the highest-scoring open-weight model in its comparison, while Claude Opus 5.5 scored 57.6. Kimi K3, the previous model in Moonshot’s series, scored 43.6. Those results give context to K3.1 Pro’s 55.1 score, but the comparison is specific to Artificial Analysis’s index and model settings; it is not a general guarantee of performance.

Moonshot has also publicised results from its own tests and demonstrations. It reported 92.7% on BenchCAD with tools, below the figures it cited for Claude Opus 5.5 and GPT-6 Astra. The company said an internal finance benchmark reached 64.4 on OfficeQA Pro v2 and reported a 65% reduction in claim-level hallucinations compared with K3 on its Kimi Hallu Bench Hard test. These are vendor-reported findings, not independent confirmation of broad, everyday performance.

Artificial Analysis’s own AA-Omniscience evaluation found a lower hallucination rate for K3.1 Pro than K3: 44% versus 53%. The outlet’s result supports a reduction on that test, but it is smaller than Moonshot’s internal figure and does not show how often the model makes errors across all real-world use. Moonshot itself says its internal hallucination test was deliberately designed to provoke errors and does not represent everyday error rates.

“The hallucination 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

Moonshot has promised to publish the K3.1 Pro weights by November 4, but they were not available at launch. The company has also not published a K3.1 licence, so it is not yet clear what downloading, modifying or commercial use will be permitted. The phrase “open-weight” alone does not settle those legal terms.

Independent benchmarks provide evidence for particular test results, not a complete account of how the model will perform in production. Moonshot’s demonstrations of game development, 3D modelling and finance work are company claims, and its internal tests have not established equivalent outcomes across ordinary user tasks. The model’s deployment costs and hardware needs for organisations seeking to run it themselves are also not specified in the material available at launch.

The November 4 Release Test

The next clear milestone is whether Moonshot publishes the weights by its stated November 4 deadline, along with the licence and technical information needed to evaluate them. If the release proceeds, independent reviewers and developers can examine whether the published model corresponds to the evaluated version and test its performance beyond the company’s demonstrations.

Until then, users can access K3.1 Pro through Moonshot’s listed apps, tools and API, subject to the provider’s terms. Readers comparing it with other models should treat the benchmark ranking as a useful snapshot, while waiting for the weight release, licence terms and further independent testing to clarify how open and practical the model will be.

Key Questions

What is Kimi K3.1 Pro?

It is a Moonshot AI model that the company describes as a more heavily post-trained version of Kimi K3. Artificial Analysis lists it as a 2.8-trillion-parameter mixture-of-experts model with 104 billion active parameters.

How did it rank on Artificial Analysis’s Intelligence Index?

K3.1 Pro scored 55.1 and ranked fourth among the models in the supplied comparison, behind Claude Opus 5.5, Claude Sonnet 5.5 and Muse Wonder Full.

Are K3.1 Pro’s weights available?

No. Moonshot had promised the weights by November 4, but they were not available at the October 23 launch. The licence had not been published either.

How much does K3.1 Pro cost through the API?

The listed price is $2 per million input tokens and $10 per million output tokens. Artificial Analysis estimated $2.11 per Intelligence Index task, a benchmark-specific cost measure.

Are Moonshot’s performance claims independently verified?

Not all of them. Artificial Analysis independently evaluated the model on its own benchmarks. Moonshot’s demonstrations and figures from its internal tests, including the 65% hallucination reduction claim, remain company-reported results.

Source: ThorstenMeyerAI.com