🔍 Read the full analysis: Kimi K3.1 Pro’s Frontier-Near Score: What’s Known And What Isn’t on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Artificial Analysis gave Moonshot AI’s Kimi K3.1 Pro an Intelligence Index score of 55.1, placing it fourth among the models listed and three points behind the highest-scoring entry. The model is available through Kimi products and API, but its weights are only promised by November 4, 2026, and no K3.1 licence has been published. Some performance and hallucination claims come from Moonshot’s own tests rather than independent evaluations.
Moonshot AI launched Kimi K3.1 Pro in its apps, Kimi Work, Kimi Code and API on October 23, and independent evaluator Artificial Analysis scored it 55.1 on its Intelligence Index, fourth among the models in its October 24 comparison. The result puts it three points behind the leading listed model, but K3.1 Pro is not yet open-weight: Moonshot says it plans to publish the weights by November 4, and has not published a licence for this version.
Artificial Analysis lists K3.1 Pro as a mixture-of-experts model with 2.8 trillion total and 104 billion active parameters, a one-million-token context window and text and image input. Its Intelligence Index score of 55.1 sits below Claude Opus 5.5 at 57.6, Claude Sonnet 5.5 at 56.0 and Muse Wonder Full at 55.8. These figures reflect the models and settings in Artificial Analysis’s comparison as of October 24, with each model evaluated at its highest reasoning setting.
The evaluator’s results show a marked gain on agentic coding tasks: K3.1 Pro scored 67% on Terminal-Bench 4.0, compared with 13% for Kimi K3 in the same table. On GDPval-AA, a simulated 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%.
Artificial Analysis lists K3.1 Pro’s cost at $2 per million input tokens and $10 per million output tokens, and estimates $2.11 per Intelligence Index task. It estimates $5.98 per task for Opus 5.5 in the same comparison. These task-cost estimates are specific to the evaluator’s benchmark and should not be treated as a universal cost for every workload. Moonshot’s product availability and pricing concern hosted access; they do not mean users can already download the model weights.
Infographic 01 · Claimed vs. measured
Moonshot says “matches the strongest.” Independent tests say: close.
The KPIs at the top are independent measurements. Vendor numbers and independent numbers come from different harnesses, so they are shown side by side, never mixed.
Moonshot reports
vendor claims own harness, partly internal tests
| Benchmark | K3.1 Pro | Comparison |
|---|---|---|
| Terminal-Bench 4.0 | 65.7% | vs. Claude Opus 5.5: 66.4% |
| BenchCAD with tools | 92.7% | vs. Claude Opus 5.5: 96.2% |
| OfficeQA Pro v2 | 64.4 | vs. Claude Opus 5.5: 62.2 |
| Kimi Hallu Bench Hard internal | 8.3% | vs. Kimi K3: 23.4% (about −65%) |
Artificial Analysis measured
independent highest reasoning setting per model
| Measurement | K3.1 Pro | Comparison |
|---|---|---|
| Terminal-Bench 4.0 | 67% | vs. Claude Opus 5.5 (max): 60% |
| GDPval-AA | 1,846 Elo | vs. Claude Opus 5.5: 1,866 |
| Humanity’s Last Exam | 52% | vs. Claude Opus 5.5: 61% |
| AA-Omniscience hallucination rate | 44% | vs. Kimi K3: 53% · Gemini 4 Argon: 15% |
The 65% hallucination drop holds on Moonshot’s own stress test. On an independent knowledge test, the rate fell from 53% to 44%.Moonshot: the stress test rates “do not represent overall error rates in daily use.”
Sources: Moonshot AI, Kimi K3.1 Pro tech blog (23 Oct 2026) (left card, vendor charts); Artificial Analysis launch article (23 Oct 2026) and Artificial Analysis model page (data as of 24 Oct 2026) (KPIs and right card).
Powered by Thorsten Meyer AI
A Smaller Gap for Open Weights
If Moonshot releases the weights as promised, K3.1 Pro could narrow the measured gap between an open-weight model and the highest-ranked proprietary models in this comparison. Artificial Analysis says it would be the leading open-weight model once its weights are published. That statement is conditional on the release; the model is currently available through Moonshot’s hosted services, not as a published weight package.
For businesses, the distinction affects control as much as benchmark position. A downloadable model may give organizations more choice about where to run it and how to adapt it, subject to its licence and hardware requirements. But a model described as open-weight is not automatically unrestricted, and the licence for K3.1 Pro is not yet public. Its listed 2.8 trillion total parameters also make self-hosting a substantial infrastructure question, rather than a simple alternative to an API subscription.
The pricing comparison is relevant to teams using hosted models, particularly for repeated agentic tasks. Still, a benchmark task-cost estimate does not establish savings for a company’s own workflow. Results can vary with prompt length, reasoning settings, tool use and the amount of output generated. The independent score provides a useful point of comparison, not a guarantee of equivalent performance or cost in production.
As an affiliate, we earn on qualifying purchases.
What the Benchmarks Actually Measure
Kimi K3.1 Pro is described in the source material as a more heavily post-trained version of Kimi K3. The earlier model scored 43.6 on Artificial Analysis’s Intelligence Index and 13% on Terminal-Bench 4.0 in the figures supplied. K3.1 Pro’s respective scores are 55.1 and 67%. The increase in the agentic coding benchmark is much larger than its change on Humanity’s Last Exam, where the reported score moved from 47% to 52%.
Moonshot has also promoted examples involving game development in Unity, Unreal Engine and Godot; parametric 3D work in Blender and CAD; and finance tasks. Those examples and several reported results are Moonshot’s own demonstrations or internal benchmarks. For instance, the company says K3.1 Pro scored 92.7% on BenchCAD with tools, compared with 96.2% for Opus 5.5 and 95.9% for GPT-6 Astra. The supplied material does not establish that this result was independently reproduced.
The hallucination figures illustrate why the testing source matters. Moonshot reports a fall in claim-level hallucinations from 23.4% for Kimi K3 to 8.3% for K3.1 Pro on its internal Kimi Hallu Bench Hard test, describing that as a 65% reduction. The company says the test was deliberately designed to provoke errors and that its rates do not represent typical daily use. Artificial Analysis measured a smaller decrease on its separate AA-Omniscience evaluation, from 53% to 44%. Different tests and conditions mean those figures are not interchangeable.
“The company says the K3.1 Pro weights will be published “by Nov 4.””
— Moonshot AI
GPU workstation for local large language models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Release and Licence Questions
The most immediate unknown is whether Moonshot publishes the weights by its stated November 4, 2026 target. The source material gives no published K3.1 Pro licence, so users cannot yet assess the terms for commercial use, redistribution, modification or deployment. Until both the weights and licence are available, claims about practical access and openness remain provisional.
It is also unclear how closely the company’s showcased projects and internal scores will translate to independent, repeatable results in ordinary use. Artificial Analysis provides independent benchmark measurements, but those tests cover particular tasks and settings; they do not verify every vendor demonstration. The supplied information also does not provide enough detail to infer how K3.1 Pro will perform across different languages, industries, tool integrations or long-running production workloads.
Finally, the model’s scale raises deployment questions. The listed 2.8 trillion total parameters describe the model’s full mixture-of-experts architecture, but the supplied material does not specify the hardware, memory, serving configuration or operating costs required for a self-hosted release. The hosted API price is not a measure of self-hosting cost.
As an affiliate, we earn on qualifying purchases.
Watch for the November Weight Release
The next stated milestone is Moonshot’s promised publication of the weights by November 4. Readers and potential adopters should look for the release itself, the accompanying licence and any technical instructions needed to run the model. Those details will determine whether organizations can use it outside Moonshot’s hosted services and under what conditions.
After release, independent evaluation of the downloadable model could help establish whether its performance matches the hosted version and Moonshot’s demonstrations. For now, K3.1 Pro’s score of 55.1 is a measured result from Artificial Analysis, while claims about specific end-to-end projects and the larger hallucination reduction remain tied to Moonshot’s own testing. The key near-term question is whether the promised weights and licence arrive, and what independent users can reproduce with them.
large language model prompt engineering books
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What is the confirmed benchmark result for Kimi K3.1 Pro?
Artificial Analysis scored it 55.1 on its Intelligence Index as of October 24, 2026, placing it fourth in the comparison described in the source material. It scored 67% on the evaluator’s Terminal-Bench 4.0 test.
Can users download Kimi K3.1 Pro now?
No. It launched through Kimi apps, Kimi Work, Kimi Code and the API on October 23. Moonshot says the weights are due by November 4, but they had not been published in the supplied reporting.
Has Moonshot published the K3.1 Pro licence?
No licence for K3.1 Pro is reported as published. The terms for commercial use, modification and redistribution therefore remain unknown.
Are the 65% fewer hallucinations claim independently verified?
No. That figure comes from Moonshot’s internal Kimi Hallu Bench Hard test. Artificial Analysis reported a decrease from 53% to 44% on its separate AA-Omniscience evaluation, which is a different measure.
Does the benchmark score prove it will be cheaper for every company?
No. Artificial Analysis estimates $2.11 per Intelligence Index task for K3.1 Pro, but that is a benchmark-specific estimate. Actual costs depend on a company’s workload, usage and settings.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
