Open-Weight Kimi K3.1 Pro Nears The Frontier Before Its Weights Release
AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Open-Weight Kimi K3.1 Pro Nears The Frontier Before Its Weights Release 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 through its apps and API on Oct. 23, and Artificial Analysis gave it an Intelligence Index score of 55.1, placing it fourth in its comparison. The model’s weights are promised by Nov. 4, but they are not yet available and Moonshot has not published a licence for them. Independent and vendor tests show strong results in some areas, while the scope of those results and the release terms remain open questions.

Moonshot AI launched Kimi K3.1 Pro on Oct. 23 through its Kimi apps, Kimi Work, Kimi Code and API, and says it will publish the model weights by Nov. 4. In an evaluation by Artificial Analysis, it scored 55.1 on the Intelligence Index, ranking fourth behind three proprietary models; the weights and a licence have not yet been released, so it is not currently available as an open-weight model.

Artificial Analysis lists K3.1 Pro as a mixture-of-experts model with 2.8 trillion total parameters and 104 billion active parameters. The evaluator also lists a one-million-token context window and text and image input. Its score of 55.1 puts it just below Claude Opus 5.5 at 57.6, Claude Sonnet 5.5 at 56.0 and Muse Wonder Full at 55.8. These are figures from the model page as of Oct. 24, with models tested at their highest reasoning settings.

The evaluator’s estimate puts K3.1 Pro at $2.11 per Intelligence Index task, compared with $5.98 for Opus 5.5 at maximum effort. The listed API rates for K3.1 Pro are $2 per million input tokens and $10 per million output tokens. These comparisons reflect the specific evaluation and pricing data cited; they do not establish that the models have identical performance across all tasks or usage patterns.

Artificial Analysis recorded a sharp gain on its Terminal-Bench 4.0 agentic-work test: K3.1 Pro scored 67%, compared with 13% for Kimi K3. It also scored 1,846 Elo on GDPval-AA, a simulated knowledge-work evaluation, behind Opus 5.5 at 1,866 and Muse Wonder Full at 1,857. On Humanity’s Last Exam it scored 52%, compared with 47% for K3 and 61% for Opus 5.5. Artificial Analysis measured K3.1 Pro’s claim-level hallucination rate at 44% on AA-Omniscience, down from 53% for K3.

At a glance
reportWhen: Launched Oct. 23, 2026; weights promise…
The developmentMoonshot AI launched Kimi K3.1 Pro and said it plans to release the model weights by Nov. 4, while independent testing placed it close to leading proprietary models.
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 Narrower Gap for Open Weights

If Moonshot publishes the weights on schedule and under terms that permit the intended uses, K3.1 Pro could bring a high-ranking model closer to researchers and organisations that want to inspect or run model weights themselves. Artificial Analysis said the model would lead open-weight systems once the weights are published. That is a conditional assessment, not a present ranking among available weights: the evaluator’s comparison listed MiMo-V2.6-Pro at 46.3 and DeepSeek V4.1 Flash at 45.1, while K3.1 Pro’s 55.1 score is for a model whose weights are still promised.

The price comparison may matter for teams choosing hosted models, particularly for agentic tasks. Still, a lower estimated cost per benchmark task does not guarantee lower costs for every workload. And downloading weights would not automatically make this model practical to run locally: its 2.8-trillion-parameter total size points to substantial infrastructure requirements, even though only 104 billion parameters are active according to Artificial Analysis.

Ninja CREAMi Scoop & Swirl Ice Cream and Soft Serve Maker, 13 Programs, 4 Pints, NC701

Ninja CREAMi Scoop & Swirl Ice Cream and Soft Serve Maker, 13 Programs, 4 Pints, NC701

As an affiliate, we earn on qualifying purchases.

From Kimi K3 to K3.1 Pro

Moonshot describes K3.1 Pro as a more heavily post-trained version of Kimi K3. The distinction matters because the launch combines vendor claims about capabilities with external benchmark results. Artificial Analysis provides the cited independent scores; Moonshot’s demonstrations and internal tests are a separate body of evidence and should not be treated as independently verified results.

Moonshot says K3.1 Pro can create projects in Unity, Unreal Engine and Godot, build parametric 3D models, and handle finance work. In one internal test, the company says the model improved a simulated human’s walking time to 11.2 seconds across 167 missions and corrected 36 errors in shoulder-muscle calculations. Moonshot also reports a 92.7% result on BenchCAD with tools, compared with 96.2% for Opus 5.5 and 95.9% for GPT-6 Astra. Those figures come from Moonshot’s reported tests, not from the Artificial Analysis comparison.

The vendor’s headline claim of 65% fewer hallucinations comes from its own Kimi Hallu Bench Hard stress test, where its reported rate fell from 23.4% for K3 to 8.3% for K3.1 Pro. Moonshot says the test is deliberately designed to provoke errors and that the results do not represent overall error rates in daily use. Artificial Analysis reported a smaller reduction on its separate AA-Omniscience evaluation, from 53% to 44%.

“The test is deliberately crafted to provoke errors.”

— Moonshot AI

Weights, Licence and Real-World Performance

The main outstanding detail is whether Moonshot will release the weights by Nov. 4, as promised. The source material says no K3.1 licence has been published, leaving unclear what users will be permitted to do with the weights if they arrive. Until the release and terms are available, the model’s open-weight status remains prospective.

There are also limits to what the published scores establish. Benchmark results do not show how the system will perform across every real-world task, and Moonshot’s product demonstrations and internal results have not been independently verified in the cited material. The available information does not specify hardware requirements, operational costs for self-hosting, or whether the full model will be practical outside data-centre-scale infrastructure.

The Nov. 4 Weight Release

The next milestone is Moonshot’s promised Nov. 4 release of K3.1 Pro’s weights. A published licence will be needed to clarify whether and how users can modify, deploy or redistribute them. If the weights appear, independent testing of the downloadable version and further details about hardware demands could help establish how closely its practical performance matches the results reported for the hosted model.

Until then, K3.1 Pro can be assessed through the company’s apps and API, while the broader claim that it is an open-weight alternative near the frontier remains dependent on the promised release. The cited benchmark standings and prices are a snapshot from Oct. 24, 2026, and may change as evaluators update their comparisons.

Key Questions

Has Moonshot released Kimi K3.1 Pro’s weights?

No. Moonshot said it plans to publish the weights by Nov. 4, 2026. The model launched through its apps and API on Oct. 23, but the weights were not yet available in the cited material.

How did Kimi K3.1 Pro rank in independent testing?

Artificial Analysis gave it an Intelligence Index score of 55.1, ranking it fourth in the comparison reported on Oct. 24. It trailed Opus 5.5, Sonnet 5.5 and Muse Wonder Full.

Is K3.1 Pro already an open-weight model?

Not yet. Its weights are still promised rather than published, and Moonshot has not published a K3.1 licence in the source material. Artificial Analysis describes its potential open-weight lead as conditional on publication.

What does Moonshot’s 65% hallucination reduction claim mean?

It refers to Moonshot’s internal Kimi Hallu Bench Hard test, where the company says the rate fell from 23.4% for K3 to 8.3% for K3.1 Pro. Moonshot cautions that the stress test does not represent overall error rates in daily use; Artificial Analysis measured a smaller fall on its separate test.

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

Global Food Trend Watch: Lahori Zeera’s Increasing Coverage

GDELT data flagged Lahori Zeera’s surge in global media coverage with an 88/100 signal score. Here’s what’s confirmed, and why food brands are watching.

A Skill Is a Folder, Not a Prompt: What Anthropic Learned Running Hundreds of Them

Anthropic reveals that Skills are folders containing instructions, scripts, and knowledge, transforming AI agent design and organizational workflows.

Can Genie 4 Build Worlds You Can Actually Explore?

Google DeepMind says Genie 4 creates interactive worlds with live text changes and sound, but access is limited and sessions remain short.

A Human-Sounding Voice Isn’t Proof: The Case Of Block’s Dessa

Block’s Dessa demos show expressive, real-time conversation, but the model has not been independently tested or publicly released.