Kimi K2.7 Code is the default planner, not a renderer

runpod/kimi-k2.7-code is the documented default on the nine chat models worth pinning. Text-only. Changing it changes the plan, not the MP4, unless you also let generate routing wander.

The agent has two catalogs. Chat models decide what to generate. Generation models decide the pixels. Kimi K2.7 Code is the brain the picker seeds from when you send nothing. It does not draw the clip.

Two catalogs, one default

Chat models live on GET /api/v1/agentic/chat-models. Each entry has id, name, description, provider. The payload also has default. Treat that list as live. Do not hard-code a roster from a blog post. As of the two-catalog write-up, default is runpod/kimi-k2.7-code (Kimi K2.7 Code). The app seeds its picker from that value and only falls back when chat_model is omitted or unknown.

Generation models live on GET /api/v1/agentic/models and on the public catalog. The agent fills a required models argument on generate tools either with a name you gave it or with a routed pick. That is a separate decision. Picking the chat model behind the agent is the rule. 9 chat models to pin behind the Versely agent is the install list.

Send the id, not the marketing name, on POST /api/v1/agentic/chat and on /chat/stream. Also send chat_model_explicit: true when the value is a choice, not a leftover default.

Custom brains are allowed. If chat_model is not in the built-in map but looks like org/model, it is treated as a custom OpenRouter model. That is the door for z-ai/glm-5.3 and x-ai/grok-4.6. GLM 5.3 is not the default. Saving “GLM” as a preference does not replace Kimi until you send or save that id.

Chat tokens and generations bill on two meters: credits_used for the reasoner, generation_credits_used for the renders. Changing brains changes the first number. It does not change what a named renderer costs. There is no free chat allowance.

Text-only is the constraint

Kimi K2.7 Code is text-only, same as the rest of the RunPod-hosted Kimi ids, the GLM 5 family, MiniMax M2.7, Qwen3 Max Thinking, the MiMo line, and DeepSeek V4 Pro. An image part 404s.

The live /chat-models list does not return a vision field. From the built-in registry, Gemini 3.1 Pro, the Claude ids, Grok 4.3, OpenRouter Kimi K2.5 / K2.6 / K3, and Qwen3.5 397B accept image parts. x-ai/grok-4.6 is the custom OpenRouter vision slug if you have saved it. It is not Grok Imagine Video.

Practical rule for the default:

• No attachments, lots of tool use. Leave Kimi. Switch if the planner is dropping arguments or over-calling tools, not because you dislike the stills. The stills are the renderer.
• The brief is a still, a frame, or a screenshot. Do not send it to Kimi. Pick a vision brain, or put the visual in text and skip the attachment.
• Video or audio attachments. The server pre-processes those through Gemini and injects an understanding into the prompt, so a text-only brain can still act. You do not have to leave Kimi just to describe a clip. You do have to leave it if the image is the brief.

RunPod Kimi and OpenRouter Kimi K3 are different ids on purpose: different endpoints, and the RunPod pair is text-only. Do not copy the name from one and send the other.

Changing the brain is not changing the clip

A prompt that keeps the knobs apart:

Use this chat model for planning. Do not change the video model. Generate with Kling 3 Turbo, 5s, 9:16, locked-off bottle on wet stone. If you need to look up allowed durations, call the schema tool, then generate. Do not substitute a different renderer.

chat_model on the request selects the planner. The named renderer in the body selects the pixels. Kling 3 Turbo is a generate row. Kimi will not make the next clip look like Kling, and Kling will not plan the tool calls.

A lasting preference is the right place to store “always plan with Kimi K2.7 Code” if you are working in the app rather than passing the field every call. It still does not pin the renderer. The agent still loads the same extras regardless of brain: brand kit, memories, media library, workflow summary. A precise brief on the default often beats a flagship brain with “make me an ad.”

When routing wanders

The claim’s second sentence is the failure mode. Changing chat_model changes the plan. It changes the MP4 only if the new brain picks a different generate model than the last one did.

Two ways that happens:

You omitted the renderer. A newly clever planner “helps” by routing to another family. You blame Kimi, or you blame GLM, for a look neither of them rendered. Name the row.

You omitted chat_model_explicit: true on an OpenRouter brain. Fast-eligible sub-agents — generation, slideshow, social, web editor — may run on Claude Haiku 4.5 on the OpenRouter path. Workflow scene-directing and movie planning keep the brain you picked. The default RunPod Kimi is not swapped for Haiku. If you stayed on runpod/kimi-k2.7-code, this stand-in does not apply. If you moved to GLM 5.3 or Grok 4.6 and skipped the flag, it might.

When you are exploring looks, let generate routing work and keep the brain stable. When you are exploring planners, keep the renderer named and swap chat_model. Kimi K2.7 Code is the documented default planner. It is text-only. It is not a renderer. Pin the MP4 separately, or accept that a wandering generate route is a second knob you left open.

FAQ

Is GLM 5.3 the default in Versely chat?

No. Treat /chat-models as live. The documented default at the time of the two-catalog post is runpod/kimi-k2.7-code. GLM 5.3 is a custom OpenRouter id: z-ai/glm-5.3.

Will changing off Kimi change my next clip’s look?

Only if the new brain picks a different generate model than the last one did. Pin the renderer in the brief — or in a saved workflow — if matching matters. The chat model changes planning, tool use, and the text reply.

Can I attach a product still to the default brain?

Not as an image part. Kimi K2.7 Code is text-only; the image part 404s. Pick a vision-capable chat model, or describe the still in text. Video and audio attachments are pre-processed for non-Gemini brains, which is a different path.

Why did a sub-agent ignore the model I picked?

If you were on an OpenRouter brain and omitted chat_model_explicit: true, a fast-eligible domain may have run on Haiku. The default RunPod Kimi is not swapped. Set the flag when the picker value is a real choice.