Moonshot AI's Kimi K3.1 Pro weights are available on Hugging Face, where the repository file listing reports a total size of 1.56 TB. The files were accessible when checked on November 4, the deadline Moonshot gave in its launch announcement. Teams considering their own deployment can now inspect the artifacts and licence, though availability alone says little about the hardware bill.
The weights follow the hosted model launch. This report examines the published files and documentation. We haven't downloaded the full model, run inference or reproduced Moonshot's benchmarks.
Active parameters don't describe the download
The model card lists 2.8 trillion total parameters, with 104 billion activated, in a mixture-of-experts architecture. It documents text and image input, a context length of 1,048,576 tokens, and MXFP4 weights with MXFP8 activations from quantization-aware training.
The distinction between total and active parameters matters for deployment planning. The active count describes how much of the model participates in computation, rather than the size of the complete artifact. The files page lists weight shards named with a 96-file sequence. A smaller active count doesn't make that repository a desktop-sized download.
Moonshot recommends vLLM, SGLang and TokenSpeed in its deployment section. That recommendation is a starting point for checking the linked recipes, not evidence that every installation or hardware configuration works. Neither the parameter count nor the repository size establishes runtime memory requirements, throughput or total operating cost.
Read the service conditions before building a business
The Kimi K3.1-Pro License permits use, modification, distribution and commercial activity, subject to conditions. Those conditions make the licence part of a deployment decision.
Section 2 requires a separate agreement with Moonshot before commercial use if the licensee or an affiliate operates a Model as a Service business and their combined revenue exceeds $20 million over any consecutive 12 months. Its definition covers access that gives another party meaningful control over model inputs, parameters or training data. It excludes capabilities embedded solely within specific product features or harnesses, and merely relaying requests to models hosted elsewhere.
Section 3 requires prominent model attribution for commercial products above its stated user or monthly revenue thresholds. Section 4 exempts internal use and access through Moonshot's official products or certified inference partners from Sections 2 and 3. These are the published terms, not a finding that any particular company's deployment qualifies for an exemption.
The release gives operators artifacts to evaluate against Moonshot's specifications. Watch for reproducible deployments that report hardware, inference-engine versions, memory use and task results together. Until those are available, the documented architecture and licence support a feasibility assessment, while the cost and quality of a particular deployment remain unmeasured here.