Availability follow-up
Checked November 4: Moonshot’s official Kimi K3.1 Pro repository publicly lists model weights, the learned numbers supplied as files for people who want to host the model themselves. We read public repository information and one file’s response headers, without downloading model files or running the model. [1] [2]
Repository metadata gives a November 2 creation date. When it first became public remains unestablished; November 4 is when we checked. [2]
Check the deployment requirements
The listing contains 96 weight files totaling about 1.56 TB. That total describes the files; the memory needed to run them requires a separate check. The model card links setup recipes for vLLM, SGLang, and TokenSpeed, software for running a model so apps can use it. [1] [2]
The repository uses a custom Kimi license. Read its conditions for your planned use; we have not assessed any reader’s eligibility. [3]
Our take
Public weights are useful for experienced self-hosting technicians because they make a deployment assessment possible. Imagine a team wanting to run a document helper on its own equipment. First compare the complete download and chosen setup recipe with the equipment available, then check the license for that use.
If those checks leave a workable plan, try a small, non-sensitive document and compare the answer with a result you already know. A repository listing cannot settle quality, speed, or the final operating cost.