Disclosure: this article contains no affiliate links. All links are plain links, and we earn nothing if you click them. See our affiliate disclosure. Based on Moonshot’s Hugging Face model card and license, Moonshot’s X post and API platform, OpenRouter’s listing and TechNode’s report, checked November 14, 2026.
AI-assisted: AI helped gather and draft this research; the facts were checked against the cited sources.
Quick answer: Moonshot AI released the full open weights for Kimi K3.1 Pro on Hugging Face on November 4, 2026 (UTC); Moonshot announced it on X at 16:00 UTC, midnight on November 5 in Malaysia. The MoE model has about 2.8 trillion total parameters, roughly 104 billion active, and a one-million-token context. Commercial use is allowed under Moonshot’s license, but large Model-as-a-Service operators above $20 million in trailing 12-month revenue need a separate agreement. Self-hosting needs serious hardware, not a laptop.
Update, November 14, 2026: the model is no longer Moonshot-only on OpenRouter. At release, OpenRouter served it from Moonshot alone, according to CellCog’s release tracker; it now lists Kimi K3.1 Pro from 18 providers (24 endpoints), with Moonshot’s own endpoint at $2.00 input / $10.00 output per 1M tokens and the lowest listed input price at $0.2375, with $8.4114 output (Relace). Moonshot’s own API lists the same $2.00 / $10.00 prices, plus a high-speed variant at $4.00 / $20.00 with the same intelligence and 1M-token context at roughly 3× the speed.
What dropped
The Hugging Face model card describes Kimi K3.1 Pro as an open-weight, natively multimodal agentic model, Moonshot’s most capable 3T-class model to date. The repo was public and ungated when we checked on November 14, 2026 (MYT).
TechNode (Nov 5) reported the weights release and noted Huawei Ascend inference support via vLLM Ascend and SGLang. An API was already available: model id kimi-k3.1-pro on platform.kimi.ai, with OpenAI/Anthropic-compatible endpoints. The card recommends vLLM, SGLang, and TokenSpeed for local or partner inference.
On Moonshot’s own API, Kimi K3.1 Pro costs $2.00 per million input tokens and $10.00 per million output tokens (cache write $2.00, cache hit $0.20). A high-speed variant with the same intelligence and 1M-token context at roughly 3× the speed costs $4.00 input and $20.00 output per million tokens.

Specs at a glance
| Spec | Value (per model card) |
|---|---|
| Architecture | Mixture-of-Experts (MoE) |
| Total parameters | 2.8T |
| Activated parameters | 104B |
| Context length | 1,048,576 tokens |
| Vision | MoonViT-V2 |
| Quantization note | MXFP4 weights / MXFP8 activations (QAT) |
| API model id | kimi-k3.1-pro |
The card lists long benchmark tables. These are Moonshot’s own reported results, not our rankings. Compare them yourself if you need a bake-off.
License in plain English
We checked the LICENSE file on the model repo:
- Most commercial use is allowed, including fine-tuning and derivatives, if you keep the copyright notice.
- If you operate a Model-as-a-Service business and you (plus affiliates) exceed $20 million USD in aggregate revenue over any consecutive 12 months, you must sign a separate Moonshot agreement before commercial use of the Software or derivatives.
- If a commercial product has more than 100 million monthly active users or more than $20 million USD monthly revenue, you must prominently display “Kimi K3.1-Pro” on the UI.
- Internal use and use through Moonshot’s official products or certified partners are carved out of those MaaS / big-product rules.
Re-read the LICENSE before you ship anything customer-facing.

Self-host reality check
This is a 2.8T MoE-class model. Quantization on the card does not mean it runs free on a consumer laptop. For most SMBs, the practical paths are:
- Call the hosted API (
kimi-k3.1-pro). - Use a certified inference partner.
- Self-host only if you already have cluster-scale GPUs and ops.
Need cheaper long-form writing rather than a self-hosted frontier MoE? A paid writing tool covered in our Agility Writer review may be simpler. Open weights and SaaS writing tools do different jobs.
FAQ
When did the weights go public?
Moonshot announced the weights on X at 16:00 UTC on November 4, 2026 (midnight, November 5, in Malaysia); TechNode covered the release on November 5.
Can I use it commercially?
Yes under the Kimi K3.1 Pro License for most cases. MaaS operators above about $20M trailing 12-month revenue need a separate Moonshot deal. Confirm the LICENSE text before launch.
Is there an API?
Yes. kimi-k3.1-pro on platform.kimi.ai.
Can I run it on a laptop?
No. Do not assume consumer hardware is enough for full weights.
What engines are recommended?
vLLM, SGLang, and TokenSpeed, per the model card.
Related: Ripple-65B Preview free open model · Tencent Hy4 free · GitHub trending AI repos Nov 5
Sources
- Hugging Face model card, moonshotai/Kimi-K3.1-Pro: https://huggingface.co/moonshotai/Kimi-K3.1-Pro (checked 2026-11-14)
- LICENSE (raw): https://huggingface.co/moonshotai/Kimi-K3.1-Pro/raw/main/LICENSE (checked 2026-11-14)
- TechNode, “Moonshot AI releases Kimi K3.1 Pro model weights” (Nov 5, 2026): https://technode.com/2026/11/05/moonshot-ai-releases-kimi-k3-1-pro-model-weights/ (checked 2026-11-14)
- Moonshot AI on X, weights release (Nov 4, 2026, 16:00 UTC): https://x.com/Kimi_Moonshot/status/2118010528123773297 (checked 2026-11-14)
- Kimi platform API: https://platform.kimi.ai (checked 2026-11-14)
- OpenRouter, Kimi K3.1 Pro listing: https://openrouter.ai/moonshotai/kimi-k3.1-pro (checked 2026-11-14)
- CellCog, “Kimi K3.1 Pro: Open Weights Out, Price and Benchmarks” (Nov 12, 2026 update): https://cellcog.ai/blog/kimi-k3-1-release-date/ (checked 2026-11-14)
Last checked: November 14, 2026 (MYT).
2 thoughts on “Kimi K3.1 Pro Open Weights Live on Hugging Face”