Versely
    Guides

    Grok 4.6 can see the still; GLM 5.3 cannot

    x-ai/grok-4.6 accepts images and PDFs. z-ai/glm-5.3 is text-only. Pin Grok when the pack shot is the brief; pin GLM when the brief is already words.

    Versely Team••5 min read

    x-ai/grok-4.6 accepts images and PDFs. z-ai/glm-5.3 is text-only. Pin Grok when the pack shot is the brief. Pin GLM when the brief is already words. Neither one is Grok Imagine. Imagine is the generate family. 4.6 is the chat brain.

    Grok 4.6 shipped 12 August 2026. SpaceXAI’s pitch is long-running agents and more ambitious interactive work, not a new video row. OpenRouter lists $2 / $6 per million tokens, $0.50 cache read, 500,000-token context, tools and JSON structured output. Inputs: text, images, files such as PDFs. Output: text. That last line is why this id belongs in Versely when the still is the contract.

    Vision is the fork

    Picking the chat model already splits the built-in roster: Gemini 3.1 Pro, Claude ids, Grok 4.3, OpenRouter Kimi K2.5 / K2.6 / K3, and Qwen3.5 397B accept image parts. The GLM 5 family, DeepSeek V4 Pro, RunPod Kimi, and several others 404 on an image part.

    Grok 4.6 sits on the vision side of that fork as a custom OpenRouter slug. GLM 5.3 sits on the text-only side. Same agent. Opposite attachment rules.

    If you attach the approved product jpeg and pin GLM 5.3, the call dies on the image part. If you attach the same jpeg and pin Grok 4.6, the planner can see the bottle, the label, the crop. Then it should still name the renderer: Seedance 2.5 for still-to-motion, Kling 3 Turbo only when you have no plate and you accept invention.

    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 chat model if the image is the brief. Grok 4.6 is that pick when you want an xAI brain rather than Gemini or Claude.

    Long-running is the other fork

    xAI trained 4.6 for staying with multi-step jobs: research, codebases, turning an idea into a working first version. Versely’s agent is a multi-step job: plan, schema lookup, generate, caption, maybe schedule. A brain that drops tool arguments mid-campaign is the wrong pin. A brain that keeps the renderer you named is the right one.

    That is not a license to omit chat_model_explicit: true. On the OpenRouter path, fast-eligible sub-agents (generation, slideshow, social, web editor) may stand in Claude Haiku 4.5 unless you mark the pick as a choice. Workflow and movie planning keep Grok if you sent Grok.

    Send:

    • chat_model: x-ai/grok-4.6
    • chat_model_explicit: true

    Prompt shape:

    Plan with Grok 4.6. The attached still is the product. Do not change the video model. Image-to-video on Seedance 2.5, 9:16, slow orbit, no new label. Do not invent a second SKU.

    Do not confuse 4.6 with Imagine

    Versely’s generate catalog already has xAI pixel rows:

    Pinning 4.6 does not select those rows. A clever planner may route to Imagine if you do not name a renderer. That is the look-shift bug. Name Seedance or Kling when the SKU has to hold. Name Imagine Video only when you want that family’s look on purpose.

    Grok 4.3 remains the built-in vision id if you have not saved 4.6. 4.6 is the August 12 OpenRouter slug. Check /chat-models before you assume the picker grew a new row.

    When to pick GLM instead

    GLM 5.3 is cheaper on OpenRouter’s list ($1.40 / $4.40 vs $2 / $6) and has a 1M context window. Reasoning is always on. If the campaign brief is a script, a scene list, a credit budget — no jpeg — GLM is the honest pin. The dedicated page is GLM 5.3 is a planner, not a renderer.

    If you need both (words and a still), do not staple GLM onto the photo. Use Grok 4.6 or Gemini for that turn. You can switch chat_model per request. The generate model can stay pinned the whole time.

    FAQ

    Is Grok 4.6 a Versely video model?

    No. It is a chat brain on OpenRouter. Video is Grok Imagine Video or another catalog row you name.

    Can Grok 4.6 replace captions?

    No. Captions are /video-editing/add-captions-to-video or overlay. The brain can ask for that tool. It is not the burn-in.

    Why not always use Grok 4.6?

    Vision and long-horizon help only when those are the job. Text-only GLM is cheaper for word briefs. The default Kimi is enough when you have no attachments and the planner is already behaving. Pick the fork, don’t collect brains.

    How do I install this in Claude Code?

    You don’t install Grok 4.6 as a Claude plugin. Install Versely skills (npx skills add AI-XLabs-Innovation/versely-skills) so Claude can submit Versely jobs. Pin x-ai/grok-4.6 inside /agent when Versely’s planner should be Grok.