Intel / article

moonshotai/Kimi-K3

Source: Simon Willison
Author: Simon Willison
Date: Published Jul 27, 2026

Key takeaways · AI-distilled

K3 is 2.8 trillion parameters and 1.56TB of weights on Hugging Face. At that size "you can download it" is close to theoretical for most teams; the practical access path is somebody else's API.
The license got stricter between generations. K2's tweak to MIT only made you display "Kimi K2" in your interface past 100M monthly users or $20M monthly revenue. K3 requires a separate signed agreement with Moonshot instead.
Note who the K3 clause actually catches: it triggers on operating a model-as-a-service business above $20M revenue over any consecutive 12 months, so providers hit it long before a company merely building products on the model does.
Seven providers were serving K3 on OpenRouter at launch, most at Moonshot's own $3 per million input and $15 per million output, which is the number any self-hosting plan has to beat.

Terms in this piece · Glossary

inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.

Why it matters

A major open-weights frontier model release with a license quirk that affects how engineers may legally deploy it.

Key quotes

“As promised earlier this month , Moonshot have released the weights for their excellent 2.8 trillion parameter Kimi K3. They're a hefty 1.56TB on Hugging Face.”

“Kimi introduced their own janky modified version of the MIT license with K2 back in July 2025.”

“To Kimi's credit, they make no attempt to describe this as an "open source" license in their own materials, consistently using the term "open weight" in its place.”

“OpenRouter is already offering K3 from 7 providers , most of which are at the same $3/million input and $15/million output as Moonshot AI themselves.”

Read the source ↗simonwillison.net