TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Moonshot AI has introduced Kimi K3.1 Pro and K3.1 Pro Highspeed, saying the models are available on Kimi.ai, Kimi Work, Kimi Code and its API. The company reports improvements on coding, 3D and finance benchmarks; full model weights are scheduled for release by Nov. 4, but the supplied announcement does not establish independent verification of its performance claims.
Moonshot AI has announced Kimi K3.1 Pro, a new model it describes as its most capable 3-trillion-parameter-class open-weight model to date, alongside K3.1 Pro Highspeed, which the company says runs two to three times faster. The models are available now through Kimi.ai, Kimi Work, Kimi Code and the Kimi API; Moonshot says it plans to release the full weights by Nov. 4.
In its announcement, Moonshot describes K3.1 Pro as a post-trained version of Kimi K3, developed with substantially more post-training compute. The company says it improved its reinforcement-learning framework, long-context inference, training efficiency and supporting infrastructure to train across hundreds of downstream domains. It lists a 1-million-token context window and highlights coding, agentic tasks, 3D creation and financial research as areas of improvement.
The company’s published results include a 65.7% score on Terminal-Bench 4.0, a coding-agent benchmark, compared with 58.2% for GPT-6 Astra and 66.4% for Claude Opus 5.5 in the figures Moonshot provided. On BenchCAD, a test of 3D reconstruction through CAD code generation, Moonshot reports 92.7% with tools, versus 96.2% for Claude Opus 5.5 and 95.9% for GPT-6 Astra. These are company-reported benchmark results, not independent findings supplied with the announcement.
Moonshot also reports gains on internal tests. K3.1 Pro scored 61.2% on its Kimi Code Bench 3.0 Long-horizon evaluation, up from 55.95% for K3. On internal finance measures, the company says it improved by 7.2 points in financial modeling and 11.6 points in financial research, and reached about 4.2 times K3’s performance on a simulated business-operations benchmark. The announcement says its internal Kimi Hallu Bench Hard claim-level hallucination rate dropped from 23.4% for K3 to 8.3% for K3.1 Pro.
Access, Speed and Model Claims
The launch gives users access to a new model through several existing Kimi products, while the Highspeed version offers a separate option for faster interactions. That may matter to developers and businesses running multi-step coding, research or document-analysis tasks, where response time and long context can affect how a workflow feels in practice.
Moonshot’s claims also put K3.1 Pro in direct comparison with named proprietary models across coding and 3D benchmarks. Those comparisons are relevant to users choosing between tools, but benchmark scores alone do not establish how a model performs on an individual’s workload. The company’s reported results should be read alongside the evaluation methods, tool settings and independent testing, which are not included in the supplied announcement.
The planned release of full weights is another significant part of the announcement. If released as promised, it would give researchers and developers access beyond the hosted services. The timing and terms will matter: the announcement gives a date but does not specify licensing conditions or other release details in the material provided.
GPU workstation for running large language models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
From Kimi K3 to K3.1 Pro
Moonshot presents K3.1 Pro as a post-trained iteration of Kimi K3, rather than a model developed from scratch. Its stated approach emphasizes scaling reinforcement learning and improving the systems used to train and evaluate models across many task environments. The company attributes its reported gains to that expanded post-training effort and related engineering work.
The release also includes product features beyond the model itself. Moonshot says its 3D Make tool is live on Kimi.ai and the Kimi app, allowing users to preview and edit models, download files prepared for 3D printing, or export videos and Live Photos. The company says K3.1 Pro can create assets using scripts for tools such as Blender and CAD software, without MCP connectors. These are descriptions of supported capabilities from the company, not independent assessments of output quality.
The supplied report focuses on selected benchmark scores and examples. It does not provide full evaluation protocols, model cards or external test results. It also does not give a year for the Nov. 4 weights-release date, so the calendar timing cannot be pinned down from the announcement alone.
“the first model in our new K3.1 series”
— Moonshot AI, in its Kimi K3.1 Pro announcement
high memory GPU for AI model inference
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Weights and Benchmark Verification
Several points remain open. Moonshot has not supplied, in the report provided, a year for the Nov. 4 weights release, the weights’ license, or whether every component needed to run the model will be included. The announcement also does not explain the test conditions behind the Highspeed version’s two-to-three-times speed claim, including the hardware, workload or comparison baseline.
The benchmark figures are attributed to Moonshot, and some are explicitly described as internal. The report does not include independent replication or enough methodology to compare all results directly. The supplied text also ends partway through its account of factual-accuracy testing, so the company’s full results and qualifications on that measure are not available here. Real-world performance, costs, availability limits and differences between the hosted model and eventual weights are also unspecified.
external SSD for AI model weights and datasets
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Nov. 4 Weights Release
The next stated milestone is Moonshot’s planned release of the full K3.1 Pro weights by Nov. 4. The company has not specified a year in the supplied announcement, so readers will need to check its official channels for the applicable date and release terms. Until then, users can try the models through Kimi.ai, Kimi Work, Kimi Code and the Kimi API, subject to the access arrangements on those services.
Further details that would help users evaluate the launch include model documentation, licensing, reproducible benchmark methods and clarification of the Highspeed comparison. Independent testing and hands-on use will help show how the reported gains translate to specific coding, 3D and financial workflows.
USB-C workstation accessories for software developers
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What are Kimi K3.1 Pro and K3.1 Pro Highspeed?
K3.1 Pro is a model Moonshot describes as a post-trained version of Kimi K3. Highspeed is a faster option that the company says delivers two to three times the speed of K3.1 Pro.
Where can users access the models?
Moonshot says both models are available through Kimi.ai, Kimi Work, Kimi Code and the Kimi API. The announcement does not set out pricing or access limits.
When will the full model weights be released?
Moonshot says the full weights will be released by Nov. 4. The supplied report does not state the year or the license under which the weights will be distributed.
Are the published benchmark results independently verified?
The scores in the announcement are reported by Moonshot, and some tests are identified as internal. The supplied material does not include independent verification or complete benchmark methods.
Source: hn
Evergreen bestsellers Picks
bestsellers
As an affiliate, we earn on qualifying purchases.
