What We Know About Kimi K3.1 Pro And Its Three-Point Gap

🔍 Read the full analysis: What We Know About Kimi K3.1 Pro And Its Three-Point Gap 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 Oct. 23, and Artificial Analysis ranked it fourth on its Intelligence Index with a score of 55.1, about three points below the top score of 57.6. The results put the model close to leading proprietary systems, but Kimi K3.1 Pro is not yet an open-weight release: Moonshot says its weights will be published by Nov. 4, and it has not published a licence for them.

Artificial Analysis lists Kimi K3.1 Pro as a mixture-of-experts model with 2.8 trillion total parameters and 104 billion active parameters. It accepts text and image input and has a stated one-million-token context window. The model is available through the Kimi apps, Kimi Work, Kimi Code and Moonshot’s API. The stated API rates are $2 per million input tokens and $10 per million output tokens.

In Artificial Analysis’s comparison, using each model’s highest reasoning setting where listed, Kimi K3.1 Pro scored 55.1. Claude Opus 5.5 scored 57.6, while Claude Sonnet 5.5 and Muse Wonder Full scored 56.0 and 55.8. The reported cost per Intelligence Index task was $2.11 for Kimi K3.1 Pro and $5.98 for Opus 5.5. Those figures apply to the benchmark’s task-cost measure, not a universal cost for every use.

Some results point to stronger performance on agent-style tasks. Artificial Analysis measured Kimi K3.1 Pro at 67% on Terminal-Bench 4.0, compared with 13% for Kimi K3. Its GDPval-AA score was 1,846 Elo, behind Opus 5.5 at 1,866 and Muse Wonder Full at 1,857. On Humanity’s Last Exam, it scored 52%, compared with Opus 5.5 at 61%. These are benchmark results, not guarantees of performance in every workplace or software project.

A Smaller Gap in Agent Benchmarks

The score puts Kimi K3.1 Pro close to the leading proprietary models on one independent evaluation, while its Terminal-Bench result shows a large gain over Kimi K3 on that test. For companies comparing models, the combination of benchmark performance and the reported API price could make it worth evaluating for coding and agentic workflows. The actual economics will depend on workloads, output volume, latency and how reliably the model completes a task.

The possible open-weight release is a separate reason the result matters. If Moonshot publishes usable weights under a licence that permits a company’s intended use, teams could have more control over deployment than they do with a hosted proprietary service. But the model’s scale—2.8 trillion total parameters—also makes self-hosting a substantial infrastructure undertaking. Until the weights and licence are available, the benchmark score does not mean organizations can independently run or inspect the model.

From Kimi K3 to K3.1 Pro

Kimi K3.1 Pro is described in the source material as a more heavily post-trained version of Kimi K3. Artificial Analysis’s listed score for Kimi K3 was 43.6, against 55.1 for K3.1 Pro. That comparison indicates a considerable difference on the index, but it does not isolate which training changes produced it or predict gains on every task.

Moonshot’s launch material also promotes demonstrations in game engines, Blender, CAD and finance. The company says the model can build projects across Unity, Unreal Engine and Godot, and reports internal gains in financial modelling and research. One company-reported hallucination test found a claim-level rate of 8.3% for K3.1 Pro versus 23.4% for K3. Moonshot describes that test as deliberately designed to provoke errors and says its results do not represent everyday error rates.

Artificial Analysis reported a smaller decline on its separate AA-Omniscience measure: 44% for K3.1 Pro versus 53% for K3. The two figures come from different tests and should not be treated as directly interchangeable. Moonshot’s demonstrations and internal benchmark results are vendor claims; Artificial Analysis’s results are an independent evaluation, though they still reflect a particular set of tests and settings.

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

— Moonshot AI

Weights and Licence Still Pending

As of Oct. 24, the weights had not been published; Moonshot’s stated target is Nov. 4. The source material does not specify what licence will govern them, whether the release will include all components needed to run the model, or whether the date will be met. Without those details, the model’s status as a practical open-weight option remains conditional.

Other questions concern reproducibility and real-world performance. The available information does not establish how Moonshot’s game, CAD and finance demonstrations perform across independent tests, or how the model compares on a broad range of everyday tasks. Benchmark rankings can also change as evaluators add models or update methods. Moonshot’s internal hallucination result and Artificial Analysis’s separate measurement differ in scope, so neither should be read as a general error rate for all use.

Watch for the November Weight Release

The next stated milestone is Nov. 4, when Moonshot says it plans to publish Kimi K3.1 Pro’s weights. A release would make it possible to examine the files, test deployment requirements and assess the accompanying licence. Until then, companies can access the model through Moonshot’s products and API, subject to the provider’s terms and pricing.

Readers should also watch for independent evaluations beyond the launch-day benchmark set. Those could clarify whether the model’s gains extend to other coding, knowledge and agentic tasks, and whether its claimed capabilities translate into reliable results outside curated demonstrations. The current record supports a strong benchmark showing; it does not yet settle the practical questions around self-hosting, licensing or performance in specific organizations.

Key Questions

What is Kimi K3.1 Pro?

Kimi K3.1 Pro is a Moonshot AI model that the company describes as a more heavily post-trained version of Kimi K3. Artificial Analysis lists 2.8 trillion total parameters, 104 billion active parameters, image and text input, and a one-million-token context window.

How close is it to the top model?

Artificial Analysis scored it 55.1, compared with 57.6 for Claude Opus 5.5, the highest score in the supplied comparison. That is a 2.5-point difference, commonly rounded to about three points.

Can users download and run it now?

Not according to the information available as of Oct. 24. Moonshot says the weights are due by Nov. 4, and no K3.1 licence has been published. The model is available through Kimi apps and services and the company’s API.

Are the claimed hallucination improvements independently confirmed?

Not in the same form as Moonshot’s internal result. The company reported a fall from 23.4% to 8.3% on its Kimi Hallu Bench Hard test. Artificial Analysis reported a decline from 53% to 44% on its different AA-Omniscience measure.

Source: ThorstenMeyerAI.com