Kimi K3.1 Pro Brings An Open-Weight Giant Close To The Frontier
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Kimi K3.1 Pro Brings An Open-Weight Giant Close To The Frontier on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

Moonshot AI released Kimi K3.1 Pro in its apps and API on Oct. 23. Artificial Analysis ranks it fourth on its Intelligence Index, but the model is not yet open-weight: Moonshot has promised weights by Nov. 4, and no licence has been published.

Moonshot AI launched Kimi K3.1 Pro on Oct. 23, and an evaluation by Artificial Analysis placed it fourth on the group’s Intelligence Index, three points behind the top-ranked model. The release is drawing attention because Moonshot says it will publish the model’s weights by Nov. 4, which could make a model with a near-frontier benchmark score available for users to run themselves. For now, however, the weights are not public, and the company has not published a licence for K3.1 Pro.

Artificial Analysis recorded a score of 55.1 on its Intelligence Index for Kimi K3.1 Pro, using the model’s highest reasoning setting. In the same comparison, Claude Opus 5.5 scored 57.6, Claude Sonnet 5.5 scored 56.0 and Muse Wonder Full scored 55.8. The figures are benchmark results, not a guarantee that the systems will perform in the same order on every task. Artificial Analysis described K3.1 Pro as set to lead open-weight models once its weights are published—a conditional assessment, not a description of what users can download today.

The model is listed as a mixture-of-experts system with 2.8 trillion total parameters and 104 billion active parameters. Artificial Analysis lists a one-million-token context window and text and image input. Moonshot made K3.1 Pro available through the Kimi apps, Kimi Work, Kimi Code and its API. Listed API prices are $2 per million input tokens and $10 per million output tokens. Artificial Analysis estimated a cost of $2.11 per Intelligence Index task, compared with $5.98 for Opus 5.5 at its maximum reasoning setting; those task-cost estimates reflect the evaluator’s test setup.

Artificial Analysis found the largest measured improvement over the earlier Kimi K3 on agentic tasks: K3.1 Pro scored 67% on Terminal-Bench 4.0, compared with 13% for K3, the highest result in the evaluator’s reported comparison. On GDPval-AA, a knowledge-work evaluation, it scored 1,846 Elo, behind Opus 5.5 at 1,866 and Muse Wonder Full at 1,857. Its result on Humanity’s Last Exam was 52%, up from K3’s 47% but below Opus 5.5’s 61%. On AA-Omniscience, Artificial Analysis measured a hallucination rate of 44%, down from 53% for K3.

At a glance
reportWhen: Launched Oct. 23, 2026; weights promise…
The developmentMoonshot AI launched Kimi K3.1 Pro and its independent benchmark results put it close to leading proprietary models, while its weights and licence remain unavailable.
Thorsten Meyer AI
Moonshot AI · 23 Oct 2026

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.

55Intelligence Index score: 4th of all models
3 ptsbehind the leader, Claude Opus 5.5 (58)
67%Terminal-Bench 4.0: highest AA has measured
$2.11per index task, vs. $5.98 for Opus 5.5 (max)

Moonshot reports

vendor claims own harness, partly internal tests

BenchmarkK3.1 ProComparison
Terminal-Bench 4.065.7%vs. Claude Opus 5.5: 66.4%
BenchCAD with tools92.7%vs. Claude Opus 5.5: 96.2%
OfficeQA Pro v264.4vs. Claude Opus 5.5: 62.2
Kimi Hallu Bench Hard internal8.3%vs. Kimi K3: 23.4% (about −65%)

Artificial Analysis measured

independent highest reasoning setting per model

MeasurementK3.1 ProComparison
Terminal-Bench 4.067%vs. Claude Opus 5.5 (max): 60%
GDPval-AA1,846 Elovs. Claude Opus 5.5: 1,866
Humanity’s Last Exam52%vs. Claude Opus 5.5: 61%
AA-Omniscience hallucination rate44%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 Near-Frontier Score With Conditions

If Moonshot publishes the weights on schedule and provides terms that permit broad use, K3.1 Pro could give developers and organizations more control over where a high-scoring model runs. That may matter for teams with specific privacy, deployment or customization requirements, and for users who want an alternative to accessing models only through a proprietary service. But the benchmark result alone does not establish that the model can be run economically or conveniently by ordinary users.

The reported price also makes the release relevant to businesses comparing model costs. Artificial Analysis put K3.1 Pro’s estimated cost per Intelligence Index task at about one-third of Opus 5.5’s under its evaluation conditions. That is a comparison of costs in one benchmark setup, not a general estimate for every workload. K3.1 Pro’s benchmark lead in agentic testing may interest developers building systems that use tools or carry out multistep tasks, but real-world performance and costs can vary with the prompt, software and amount of reasoning requested.

There is a practical difference between open weights and an easy-to-run model. At 2.8 trillion total parameters, K3.1 Pro would require substantial computing resources to host locally, even though the mixture-of-experts design activates a smaller portion of its parameters for a given task. Publication would give users access to the model’s weights; it would not by itself establish that a typical company can operate it on its existing hardware.

De'Longhi Rivelia Automatic Espresso Machine with Milk Frother

De'Longhi Rivelia Automatic Espresso Machine with Milk Frother

Bose QuietComfort Ultra Wireless Noise Cancelling Earbuds, 2nd Gen, with 20W 2-Port Wall Charger

Bose QuietComfort Ultra Wireless Noise Cancelling Earbuds, 2nd Gen, with 20W 2-Port Wall Charger

As an affiliate, we earn on qualifying purchases.

From Kimi K3 to K3.1 Pro

Kimi K3.1 Pro is described as a more heavily post-trained version of Kimi K3. The earlier model scored 43.6 on the Artificial Analysis Intelligence Index, compared with 55.1 for K3.1 Pro in the evaluator’s latest comparison. The index aggregates performance across multiple evaluations, so the overall score captures a range of tasks rather than a single measure of capability. Individual results differ: the biggest gap cited in the source material is on Terminal-Bench 4.0, while the improvement on Humanity’s Last Exam is smaller.

Moonshot has also promoted demonstrations involving game development in Unity, Unreal Engine and Godot, parametric modelling in Blender and CAD, and finance tasks. Those examples and most of the specific results Moonshot publicized come from the company’s own demonstrations or internal testing, not from Artificial Analysis. For example, Moonshot reported a 92.7% score on BenchCAD with tools, compared with 96.2% for Opus 5.5 and 95.9% for GPT-6 Astra. The company also reported an internal 65% reduction in claim-level hallucinations against K3 on its Kimi Hallu Bench Hard test. Artificial Analysis’s separate AA-Omniscience evaluation showed a smaller change, from 53% to 44%.

Moonshot cautioned that its hallucination test is deliberately designed to provoke errors and that its rates do not represent overall error rates in daily use. The difference between the company’s internal result and the independent evaluation reflects distinct tests and should not be treated as a direct contradiction or a single settled measure of everyday accuracy.

“The test is “deliberately crafted” to provoke errors.”

— Moonshot AI

Weights, Licence and Real-World Results

The most immediate uncertainty is whether Moonshot will publish the weights by Nov. 4 as promised. The company has not published a K3.1 Pro licence, so prospective users cannot yet determine what use, modification or redistribution would be allowed. Until both the weights and terms are available, describing the model as open-weight is premature; it is a planned release rather than a currently downloadable one.

Other open questions include the compute and infrastructure required to serve the model, its performance on tasks outside the published evaluations, and how its API costs compare for particular workloads. Moonshot’s demonstrations of software and finance work are company claims, while the independent benchmark results cover selected test sets. Neither set of results alone establishes reliability across everyday professional use. The source material also does not establish how the model’s one-million-token context window performs on every long-document task.

The Nov. 4 Weights Deadline

The next milestone is Moonshot’s stated plan to publish K3.1 Pro’s weights by Nov. 4. Users and developers will then be able to check whether the release arrives as promised, review the licence, and assess what hardware is needed to run the model. Until those details are available, the benchmark ranking and API access are the clearest ways to evaluate the launch.

Further independent testing can help establish whether the gains on agentic tasks and the lower measured hallucination rate hold across additional workloads. For organizations weighing deployment, the practical comparison will depend not only on benchmark scores and listed API prices, but also on hosting costs, performance, reliability and the rights granted by the eventual licence. Those details remain pending.

Key Questions

What is Kimi K3.1 Pro?

Kimi K3.1 Pro is a model launched by Moonshot AI on Oct. 23, 2026. Artificial Analysis lists it as a mixture-of-experts model with 2.8 trillion total parameters, 104 billion active parameters, a one-million-token context window, and text and image input.

Is Kimi K3.1 Pro open-weight now?

No. Moonshot has promised to publish the weights by Nov. 4, but they were not public at the time covered by the report. The company had also not published a licence for K3.1 Pro.

How does it compare with leading proprietary models?

Artificial Analysis scored K3.1 Pro at 55.1 on its Intelligence Index, placing it fourth in the comparison. Opus 5.5 scored 57.6, Sonnet 5.5 scored 56.0 and Muse Wonder Full scored 55.8. These results cover the evaluator’s tests and do not establish that one model is best for every task.

What does the model cost to use?

Moonshot’s listed API rates are $2 per million input tokens and $10 per million output tokens. Artificial Analysis estimated $2.11 per Intelligence Index task for K3.1 Pro in its testing. Actual costs can differ by workload and usage.

What remains unknown about the launch?

The weights, licence and release timing remain unconfirmed in practice until publication. The hardware needed to run the model and how well its benchmark results generalize to everyday work also remain open questions.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
You May Also Like

Capital: The Lever Beneath the Levers

Analysis of how the flow and funding of AI infrastructure reveal the fragility of the capital chokepoint in 2026.

Unlocking Gaming’s Potential: 9 AI Technologies Leading The Charge In 2026

Discover the nine AI innovations leading gaming’s evolution in 2026, shaping immersive experiences, player engagement, and industry growth.

Grok 4.6: SpaceXAI’s Long-Context AI Model For Coding, Automation, And Persistent Agents

SpaceXAI’s Grok 4.6 introduces a 500K context window for long-running agents, coding, and knowledge work, but technical details and availability remain unclear.

How To Find Citation Gaps In SaaS Documentation

An IdeaNavigator AI proposal would check whether AI answers misstate SaaS prices, plans or features—and whether vendors will pay to track the errors.