Skip to content

Models

The plugin ships an advisory catalog of the four Kimi Code models. The catalog drives model discovery; requests to the endpoint are not restricted by it, so you can call any model id the endpoint accepts.

Model id Context Reasoning effort Image input
k3 1,048,576 low / high / max yes
k3-256k 262,144 low / high / max yes
kimi-for-coding 262,144 always on (no effort levels) yes
kimi-for-coding-highspeed 262,144 always on (no effort levels) yes

Reasoning

  • K3 family (k3, k3-256k): exposes low, high, and max reasoning effort. The default is low; override per request or with the reasoningEffort config field.
  • Coding models (kimi-for-coding, kimi-for-coding-highspeed): thinking is always on with no effort selector. The adapter never sends a reasoning_effort field to them.

Image input

All four models accept image content. The adapter advertises text and image input modalities for a model whose catalog entry sets supportsImage. See Image input.

Selecting a model

Select provider kimi-code and the model id when you run a task. For the largest context window, use k3.

Overriding the catalog

Set the models field in cordis.yml to change the advertised catalog. Each entry may set contextWindow, maxTokens, supportsReasoning, and supportsImage. A model marked without image support rejects image content rather than silently dropping it.