Skip to main content
Text modelsMoonshot AI

Kimi-K3: model overview

Kimi K3 is an open-weight multimodal model from Moonshot AI. It accepts text, images, and video. The developer specifies a million-token context, 2.8 trillion total parameters, and 104 billion active parameters. The described agent workflows require a separate tool environment.

External model reference page. This page does not confirm availability in Neiron.

Content updated:

Capabilities

Analyze written and visual material together.

Work with long technical briefs and code.

Plan actions in a separately connected agent environment.

Use cases

Use Kimi-K3 when you need repository work, code repair, or a multi-step engineering task.Compare Kimi-K3 with related Kimi models if the task allows a different input or output format.Before starting, confirm the input format and expected result (text, code, or an implementation plan).

Reviewed facts

Developer
Moonshot AI
Purpose
Language model, Vision, Coding, Reasoning, Agentic workflows
Input
Text, Images, Video
Output
Text
Verified
2026-09-15

Benchmarks

No comparable benchmark is recorded in this card.

Sources

What to verify before use

Kimi K3 has its own license. Do not automatically apply Kimi K2.5 terms to K3.
The stated context does not establish a third-party service’s limits or give the model access to a user repository.
Open-weight context in the source: 1,000,000 tokens. This is not a Neiron chat limit.
Weights license: Kimi K3 License. Check the developer repository for the applicable terms.

Prompts

These are example briefs for your own evaluation, not test results. A reference page does not imply Neiron access to the model.

Prompt example

Compare this demo recording with the notes. List differences between claimed and visible feature behavior and identify the source for every observation.

Prompt example

Review this requirements document. Separate shared constraints, exceptions, and conflicts; do not silently combine different rule versions.

Prompt example

Propose an investigation sequence from this event log. First identify missing data, then give minimal read-only checks.

FAQ

What can Kimi-K3 be used for?

Kimi K3 is an open-weight multimodal model from Moonshot AI. It accepts text, images, and video. The developer specifies a million-token context, 2.8 trillion total parameters, and 104 billion active parameters. The described agent workflows require a separate tool environment.

How do I distinguish variants and capabilities?

An agent environment's tools are not part of every Kimi chat. A large context window does not guarantee every detail is noticed; verify references to the original passages.

How should I compare outputs on my own task?

Use the same source material and prompt. Write down which facts or details must be preserved. The examples on this page are evaluation briefs, not results of a Neiron test.