9 chat models to pin behind the Versely agent in 2026

The brain plans. The catalog renders. Pin z-ai/glm-5.3 or x-ai/grok-4.6 on the agent, then name the video model in the brief. Mixing those knobs is how a new chat model “changes the look” when a router quietly picked a different renderer.

This is the install list. Each item is an id you send as chat_model, or a custom OpenRouter slug in org/model form. The live roster is GET /api/v1/agentic/chat-models. Do not hard-code this page as the picker. Picking the chat model behind the agent is the two-catalog rule; this page is nine brains and the one job each is for.

If you live in Claude Code, Cursor, or Codex, the host-tool install is separate: npx skills add AI-XLabs-Innovation/versely-skills on /skills, or paste https://mcp.versely.studio/mcp on /mcp. Skills call Versely. The chat model is still the planner inside the app.

The two catalogs

Chat models decide what to generate. Generation models decide the pixels. Swap the first and the agent plans, talks, and picks tools differently. The MP4 still comes from whichever row the generate tool was told to use.

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 GLM 5.3 and Grok 4.6 this week. Duplicate ids (a custom row that repeats a built-in) are dropped.

Chat tokens and generations bill on two meters. Changing brains changes the reasoner line. It does not change what a named renderer costs. There is no free chat allowance.

The nine

1. Kimi K2.7 Code — runpod/kimi-k2.7-code

The current default. The app seeds the picker from default on /chat-models. Text-only. Use it when you have no attachments and a lot of tool use. Switch if the planner is dropping arguments, not because you dislike the stills. The stills are the renderer.

2. GLM 5.3 — z-ai/glm-5.3

Z.ai’s GLM-5.3 is text in, text out, 1,048,576-token context, tools on. OpenRouter list price is $1.40 / $4.40 per million tokens. Reasoning is always on (max default). Same GLM-5 family rule as the built-in note: do not attach a still. The image part 404s. Pin this for long-horizon plan-and-call loops. Name Kling or Seedance in the same message. Dedicated post: GLM 5.3 is a planner, not a renderer.

3. GLM 5.1 — z-ai/glm-5.1

The older cheap orchestrator. Still on OpenRouter. Same text-only family. If a saved preference still says “GLM 5.1,” that is this id, not 5.3. Upgrade the slug when you want 5.3’s longer context and always-on reasoning. Do not assume the picker renamed itself.

4. Grok 4.6 — x-ai/grok-4.6

Grok 4.6 shipped 12 August 2026. OpenRouter: $2 / $6 per million, 500K context, tools, structured JSON. Accepts text, images, and PDFs. Returns text. That is the still-as-brief brain. Long-running agents are the training pitch, not a second generate family. Dedicated post: Grok 4.6 can see the still; GLM 5.3 cannot.

5. Grok 4.3 — built-in vision id

On the built-in registry, Grok 4.3 accepts image parts. Use it when the attachment is the brief and you have not saved x-ai/grok-4.6 yet. It is not Grok Imagine Video. Chat vs generate. The Imagine rows are pixels.

6. Claude (OpenRouter ids)

Brand voice, scripts, long captions. Claude ids are served via OpenRouter. Vision-capable. If the job is word choice on a locked still, pick Claude and pin the renderer. Do not ask Claude to invent the product.

7. Gemini 3.1 Pro

Native Google entry. Vision. Video and audio attachments are pre-processed through Gemini for non-Gemini brains; you do not have to pick Gemini just to describe a clip. You do have to pick a vision model if the image is the brief. Gemini is the native path when the attachment is the work.

8. DeepSeek V4 Pro

Text-only. Same attachment rule as GLM: no stills on the wire. Cost-efficient planning when the brief is already words. For the open-weight story see DeepSeek V4; this row is the planner id, not a video model.

9. Qwen3.5 397B

Vision-capable on the built-in registry. Use it when you need image parts and you want a Qwen brain rather than Gemini or Grok. Still not the Qwen image generate row. Two catalogs.

How to pin without changing the look

Send chat_model as the id. Send chat_model_explicit: true when the value is a choice. On the OpenRouter path, fast-eligible sub-agents (generation, slideshow, social, web editor) may stand in Claude Haiku 4.5 unless that flag is set. Workflow and movie planning keep the brain you picked.

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. If you need allowed durations, call the schema tool, then generate.

A lasting preference is the place to store “always plan with GLM 5.3” in the app. It still does not pin the renderer.

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.

Can I type “Grok 4.6” in the box?

Send the id x-ai/grok-4.6, not the marketing name. Custom OpenRouter rows show up on /chat-models after you save them.

Will pinning Grok 4.6 make my next clip look like Grok Imagine?

Only if the new brain picks a Grok Imagine generate row. Pin Kling 3 Turbo or Seedance 2.5 in the brief if matching matters.

Do I need this if I only use Claude Code skills?

Skills and MCP submit jobs through Versely. The in-app agent still has a chat model. Install the skills for the host tool. Pin the brain when you are in /agent.