Moonshot AI · llm model

Kimi K3

Release Ruler for Kimi K3: released Jul 16, 2026, $3 per 1M input tokens, open weights. Each thin stem on the ruler is another model we track.

Kimi K3 is a Moonshot AI language model released Jul 16, 2026; Moonshot AI's newest language model in our dataset is Kimi K3.1 Pro. Open weights (Kimi K3 License (custom)). Moonshot's 2.8-trillion-parameter open-weights model with native vision and a 1M-token context, released under a custom Kimi K3 license. Verified against vendor sources on Oct 26, 2026.

Reported Last verified · All models · JSON · Markdown

  • ReleasedJul 16, 2026
  • Context window1M1,000,000 tokens
  • LicenceOpen weights, restrictedKimi K3 License (custom)

Modalities not verified

Key facts

Kimi K3: key facts with sources
FactValueSource
VendorMoonshot AIn/a
ReleasedJul 16, 2026 Announcement date; weights followed on Hugging Face.source
Statusgenerally availablen/a
API idkimi-k3source
Context window1,000,000 tokens (1M)source
Parameters2.8T total, 16 of 896 experts active, mixture-of-expertssource
LicenceKimi K3 License (custom) (open weights) Model-as-a-service businesses above US$20M revenue in 12 months need a separate agreement; large products must display "Kimi K3".source
Pricing$3 in / $15 out per 1M tokens Cache-miss input $3.00, cache-hit $0.30, output $15.00.source

Where Kimi K3 sits in our data

How Kimi K3 differs from Kimi K3.1 Pro

Read from the recorded facts of the two models, not from a head-to-head test. We have not run either.

  • Listed price is $3 in and $15 out, against $2 in and $10 out for Kimi K3.1 Pro (per 1M tokens).
  • Context window is 1M tokens against 1.05M for Kimi K3.1 Pro.
  • Kimi K3 came out 99 days before Kimi K3.1 Pro.
  • Facts for the other model: Kimi K3.1 Pro.

What we tested

Nothing yet. We have not run Kimi K3 ourselves, so every fact on this page is Reported from the sources below. When we do test it, the scope will be stated here and in our test logs. See methodology.

Not verified, left out

We could not confirm these on a vendor page, so we do not state them.

  • Input/output modalities beyond "native vision" were not captured.

Sources

More from Moonshot AI

All 88 models