🔍 Read the full analysis: Three Points From Frontier Performance: Kimi K3.1 Pro Explained on ThorstenMeyerAI.com
TL;DR
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 listed. Moonshot says it will publish the model weights by November 4, but no K3.1 Pro licence has been published; performance claims beyond independent tests remain vendor-reported.
Moonshot AI launched Kimi K3.1 Pro on October 23, describing it as its most capable 3-trillion-class open-weight model, while independent evaluator Artificial Analysis scored it 55.1 on its Intelligence Index, fourth among the models in its comparison. The release is available through Kimi products and the API, but the model weights are not yet public: Moonshot has promised them by November 4, and has not published a licence for K3.1 Pro.
Artificial Analysis lists K3.1 Pro as a mixture-of-experts model with 2.8 trillion total parameters and 104 billion active parameters, a one-million-token context window and text and image input. Its Intelligence Index score of 55.1 places it behind Claude Opus 5.5 at 57.6, Claude Sonnet 5.5 at 56.0 and Muse Wonder Full at 55.8. The comparison uses each model at its highest listed reasoning setting.
The evaluator reported an estimated $2.11 per Intelligence Index task for K3.1 Pro, compared with $5.98 for Opus 5.5. Moonshot’s listed API prices are $2 per million input tokens and $10 per million output tokens. These figures reflect the evaluator’s test setup and listed prices; actual costs depend on usage and task mix.
Artificial Analysis found a particularly large difference on Terminal-Bench 4.0, a test of agentic computer tasks: K3.1 Pro scored 67%, compared with 13% for Kimi K3. Its Humanity’s Last Exam score was 52%, against 47% for K3 and 61% for Opus 5.5. On the evaluator’s AA-Omniscience test, its hallucination rate was 44%, down from 53% for K3. These results describe specific benchmarks, not guaranteed performance in everyday use.
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 Narrower Gap on Agentic Tasks
The result places K3.1 Pro close to leading closed models in Artificial Analysis’s overall ranking, and its Terminal-Bench score marks a substantial jump over the earlier Kimi K3 result on that test. For companies comparing models, the combination of benchmark performance and listed API pricing could make it a candidate for trials involving coding or multi-step tasks. The published results do not establish that it will be cheaper or better for every workload.
The open-weight question remains central to the launch’s significance. If Moonshot releases usable weights on the stated schedule, organizations may be able to examine or run the model outside Moonshot’s hosted products, subject to the licence. But weights are not available yet, and the absence of a published licence leaves important limits on use and redistribution unknown. The listed 2.8-trillion total parameter count also means self-hosting would require substantial infrastructure; open weights do not by themselves make deployment practical for every organization.
Popular right now

De'Longhi Magnifica Duo Fully Automatic Espresso Machine with Milk Frother, Silver

Ninja Luxe Café Mini Plus Espresso and Coffee Maker with AutoFroth, Bone & Champagne

Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Bluetooth 5.3, USB-C Charging, and IPX4 Water Resistance
As an affiliate, we earn on qualifying purchases.
From Kimi K3 to K3.1 Pro
K3.1 Pro is described as a more heavily post-trained version of Kimi K3. Artificial Analysis’s figures put the earlier Kimi K3 at 43.6 on the Intelligence Index, while K3.1 Pro scored 55.1. The same evaluator placed Xiaomi’s MiMo-V2.6-Pro, then the highest-ranked available open-weights model in the supplied comparison, at 46.3. These results show the scale of the benchmark change, but they do not establish equivalent performance across all uses.
Moonshot’s launch material also promotes work in game engines, 3D modeling, CAD and finance. It reports examples such as creating projects in Unity, Unreal Engine and Godot, and gives a 92.7% result on BenchCAD with tools. Those are company-reported demos and results, rather than independently verified outcomes in the material provided. On hallucinations, Moonshot says an internal stress test showed a fall from 23.4% for K3 to 8.3% for K3.1 Pro, or about 65%. Artificial Analysis measured a smaller change on its separate test, from 53% to 44%. The tests use different methods and should not be treated as direct equivalents.
„The internal 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
K3.1 Pro’s weights have not been released, and the source material says Moonshot has not published its licence. The November 4 date is a company promise, not confirmation that the files will be available then or what terms will govern them. Until that information appears, organizations cannot assess the exact permissions or practical requirements for self-hosting.
It is also unclear how the benchmark scores translate to specific customer workloads. Artificial Analysis provides independent measurements on named tests, while Moonshot’s claims about full game projects, CAD work and finance results largely come from its own demonstrations and internal evaluations. Neither set of results establishes performance across all real-world tasks. The supplied source material also ends partway through its discussion of the AA-Omniscience comparison, so it does not provide a complete account of every model’s result on that test.
The November Weights Deadline
The next stated milestone is November 4, 2026, when Moonshot says it plans to publish the weights. Readers and potential users will then be able to check whether the release arrives, review the licence, and assess the files and hardware requirements. Until then, K3.1 Pro can be accessed through the Kimi apps, Kimi Work, Kimi Code and the Kimi API, according to the launch information.
Further scrutiny will depend on public access to the model and additional testing. Independent evaluations across more tasks, alongside clear documentation of the licence and deployment requirements, would help users judge whether the launch’s benchmark advantages carry over to their own work.
Key Questions
What is Kimi K3.1 Pro?
Kimi K3.1 Pro is a Moonshot AI model described as a more heavily post-trained version of Kimi K3. Artificial Analysis lists 2.8 trillion total parameters, 104 billion active parameters and a one-million-token context window.
How did it rank in independent testing?
Artificial Analysis gave it a 55.1 Intelligence Index score, placing it fourth among the models in the supplied comparison, behind Opus 5.5, Sonnet 5.5 and Muse Wonder Full.
Can users download and run the weights now?
No. Moonshot has said the weights are due by November 4, 2026. The source material says they have not yet been released and that a K3.1 Pro licence has not been published.
What does Moonshot claim about hallucinations?
Moonshot reports a reduction from 23.4% to 8.3% on its internal Kimi Hallu Bench Hard test, describing that as about 65%. Artificial Analysis measured a change from 53% to 44% on a separate test. The results use different evaluations, and Moonshot says its internal test does not represent everyday error rates.
Source: ThorstenMeyerAI.com