AIThis post was created with the assistance of artificial intelligence (AI).

TL;DR

Moonshot AI has introduced Kimi K3.1 Pro and K3.1 Pro Highspeed, offering the models through Kimi.ai, Kimi Work, Kimi Code and its API. The company reports gains on coding, 3D and finance benchmarks; those results are vendor-reported, and full model weights are scheduled for release by Nov. 4.

Moonshot AI has announced Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed, making both available through Kimi.ai, Kimi Work, Kimi Code and the Kimi API. The company describes K3.1 Pro as its most capable 3T-class open-weight model to date, with a 1-million-token context window, and says the Highspeed version runs two to three times faster; release of the full model weights is scheduled for Nov. 4.

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

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

Moonshot says K3.1 Pro is a post-trained version of Kimi K3, developed with substantially more post-training compute. The company says it upgraded its reinforcement-learning framework, long-context inference, training efficiency and related infrastructure to support training across hundreds of downstream domains. Those descriptions come from the company; the announcement does not provide an independent evaluation of the training process.

The release highlights results on several benchmarks. Moonshot reports a 65.7% score on Terminal-Bench 4.0, compared with 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5. It also reports 92.7% on BenchCAD with tools, against 96.2% for Claude Opus 5.5 and 95.9% for GPT-6 Astra. The company describes Kimi Code Bench 3.0 Long-horizon as an internal benchmark and says K3.1 Pro scored 61.2%, up from 55.95% for K3.

For finance tasks, Moonshot reports a score of 64.4 on OfficeQA Pro v2 and gains on its internal Financial Modeling, Financial Research and Business Operation benchmarks. It says the claim-level hallucination rate on its internal Kimi Hallu Bench Hard fell from 23.4% for K3 to 8.3% for K3.1 Pro. The source material cuts off before giving the related figure for responses containing at least one factual error, so that measure cannot be reported here.

At a glance
announcementWhen: Announced as available today; the compa…
The developmentMoonshot AI announced Kimi K3.1 Pro and a faster Highspeed version, with access available now and full model weights promised by Nov. 4.

Access, speed and model openness

The announcement combines immediate product access with a later open-weight release. Users can try the models through Moonshot’s named services now, while developers and researchers seeking the model weights will have to wait until the stated Nov. 4 release date. The company does not specify in the announcement what license or usage conditions will apply to those weights.

The Highspeed option is presented as a way to reduce wait time, with Moonshot claiming two to three times the speed of K3.1 Pro. The source does not define how speed was measured, which workloads were tested, or whether the comparison applies across all supported products. That makes the claim useful as a product description, but not a fully specified performance comparison.

Benchmark results may help prospective users judge whether the model suits coding, 3D design or financial research. However, the comparisons are presented by the model’s maker, and some tests are internal. Readers should treat them as company-reported results, not independent confirmation of performance in everyday use.

Products Worth Considering

Apple AirPods Pro (2nd Generation) Wireless Earbuds with USB-C MagSafe Charging Case

Apple AirPods Pro (2nd Generation) Wireless Earbuds with USB-C MagSafe Charging Case

Apple MacBook Air 15-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Midnight

Apple MacBook Air 15-inch Laptop with M5 Chip, 16GB Memory, 512GB SSD, Midnight

Kindle Paperwhite Signature Edition 32GB (2024 Model) - Metallic Black

Kindle Paperwhite Signature Edition 32GB (2024 Model) – Metallic Black

As an affiliate, we earn on qualifying purchases.

K3.1 builds on Kimi K3

Moonshot positions K3.1 Pro as the first model in a new K3.1 series and as a post-trained successor to K3. Its announcement focuses on tasks requiring multiple steps, tool use or long inputs, including software development, 3D modeling and financial-document analysis. The stated one-million-token context window is intended to support long materials, though the release does not provide details about practical limits or performance at that length.

The company also describes an internal coding test in which the model built and refined a human-walking simulation over 167 missions and 420 independent reviews. Moonshot says the simulated figure eventually walked continuously for 11.2 seconds after the model identified and corrected 36 errors in shoulder-muscle calculations. This is an example reported by the developer, not an independently reproduced result.

Alongside the models, Moonshot says its 3D Make feature is live on kimi.ai and the Kimi app. The company says users can preview and edit generated models, download them for 3D printing, or export them as videos or Live Photos.

“Kimi K3.1 Pro is “our most capable 3T-class open-weight model to date.””

— Moonshot AI

Benchmark methods and release terms

The announcement does not provide independent testing, detailed benchmark methodology for every result, or enough information to reproduce all comparisons. In particular, Kimi Code Bench 3.0 Long-horizon and the cited finance benchmarks are described as internal tests. Results on those tests may not translate directly to other workloads.

Several practical details remain unspecified: the year attached to the Nov. 4 weights deadline, the license and access terms for the weights, and the conditions behind the Highspeed version’s claimed speed advantage. The supplied source also ends mid-sentence while describing factual-error rates, leaving that additional accuracy measure incomplete. No pricing, usage limits or separate model specifications are included in the provided material.

Weights set for Nov. 4

Moonshot says the complete K3.1 Pro weights will be released by Nov. 4. The company has not stated the release year in the supplied announcement, so the deadline’s timing cannot be pinned down more precisely from this material. Users can access the two versions through Kimi.ai, Kimi Work, Kimi Code and the Kimi API in the meantime.

Further details to watch for include the weights’ license, technical documentation and independent evaluations of the published benchmark claims. Moonshot has not announced a separate release schedule for additional K3.1 models.

Key Questions

What did Moonshot announce?

Moonshot introduced Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed. It says both are available through Kimi.ai, Kimi Work, Kimi Code and the Kimi API.

How is the Highspeed version different?

Moonshot claims K3.1 Pro Highspeed is two to three times faster than K3.1 Pro. The announcement does not explain the measurement method or identify the workloads used for that comparison.

When will the full model weights be released?

The company says the full weights will be released by Nov. 4, but the supplied announcement does not specify the year or the weights’ license terms.

Are the benchmark results independently verified?

The figures in the announcement are reported by Moonshot. Some cited tests are internal, and the supplied material does not describe independent verification of the results.

Source: hn

You May Also Like

Cursor Introduces Composer 2.5

Cursor announced the release of Composer 2.5, featuring improved intelligence, behavior, and training methods, marking a significant upgrade over Composer 2.

Gemini 3.6 Flash, 3.5 Flash-Lite, And 3.5 Flash Cyber

Google announces Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber, marking new AI model updates with enhanced capabilities for developers and users.

GPT-6 Astra

OpenAI announces GPT-6 Astra, a new AI model, with details available on deployment safety. The development signals ongoing AI advancements amid rising interest.

Could Multimodal AI Emerge In Just Two Years? Insights From SenseTime Scientist

A SenseTime researcher forecasts a major leap in multimodal AI capabilities by 2027, signaling rapid progress in the field amid global competition.