GLM-4.6V

What is GLM-4.6V?

GLM-4.6V is a family of vision-language models from Z.ai. The full GLM-4.6V is a 106B mixture-of-experts model with 12B active parameters, built for cloud servers; GLM-4.6V-Flash is a 9B model for local use, and GLM-4.6V-FlashX is a low-cost, faster option on Z.ai's API. All three read video, images, text and files, answer in text, and hold 128K tokens. They can pass images straight into tool calls and read images that tools send back.

It is no longer Z.ai's newest vision model. Z.ai still sells it, but has stopped maintaining its GitHub repository and now leads with GLM-5.3-Flash, which costs less per token and holds 1M tokens.

Is GLM-4.6V still available?

Yes. Z.ai still lists all three versions on its API, but GLM-5.3-Flash is now its newer multimodal model.

API

The model IDs glm-4.6v, glm-4.6v-flashx and glm-4.6v-flash are still on Z.ai's pricing page and in its API reference.

Open weights

GLM-4.6V and GLM-4.6V-Flash stay on Hugging Face under MIT, but Z.ai no longer maintains the GLM-V GitHub repository.

Successor

GLM-5.3-Flash takes video, images, text and files with a 1M-token context, for $0.15 / $0.50 per million tokens.

Z.ai's own tools

Its Vision MCP Server for GLM Coding Plan users now runs on GLM-5.3-Flash.

Who is GLM-4.6V best for?

GLM-4.6V is best for developers building apps and agents that read long documents, screenshots and video, from free prototypes on GLM-4.6V-Flash to self-hosted production.

• Long reports and slide decks

  The 128K context takes about 150 pages of dense documents, 200 slides or an hour of video in one request, charts and tables included.

• Screenshot to front-end code

  It turns a screenshot or design file into HTML, CSS and JavaScript, and fixes the code when you circle part of a rendered page and say what to change.

• Agents that work with images

  Function calling accepts screenshots and document pages as tool inputs and reads charts or images that tools send back, without converting them to text first.

• Free or self-hosted vision

  GLM-4.6V-Flash costs nothing on the API, and its 9B open weights are small enough to run on your own hardware.

What's included with GLM-4.6V?

Tool calls that take and return images, a thinking switch, up to 150 images or two videos per request, object locations, visual web search and MIT-licensed weights.

Tool calls with images

Screenshots and document pages go straight into tool calls, and it reads the charts or pages that tools return.

Thinking switch

You can turn its thinking mode on or off per request.

Large inputs

Up to 150 images, two videos of up to 200 MB, or PDF, Word, Excel and PowerPoint files of up to 50 MB in a request.

Object locations

It can return bounding-box coordinates for things it finds in an image.

Visual web search

It plans and calls search tools, then writes an illustrated report from the results; Z.ai's built-in Web Search tool costs $0.01 per use.

Open weights

GLM-4.6V, an FP8 version and GLM-4.6V-Flash are on Hugging Face and ModelScope under MIT, served with vLLM or SGLang.

How much does GLM-4.6V cost?

GLM-4.6V costs $0.3 per million input tokens and $0.9 per million output tokens on Z.ai's API, the lighter GLM-4.6V-FlashX costs $0.04 and $0.4, and GLM-4.6V-Flash is free.

All prices are in US dollars. The built-in web search tool costs $0.01 per use, and the older GLM-4.5V costs $0.6 / $1.8 per million input / output tokens. The weights are also free to download under the MIT license if you would rather host the models yourself. Full pricing on docs.z.ai. We checked these prices on 28 Sep 2026.

How do readers rate GLM-4.6V?

No one has rated GLM-4.6V yet. Be the first.

GLM-4.6V pros and cons

It has a free tier and open weights that run on ordinary hardware, but it can loop, is weaker on text-only work, and Z.ai's newer GLM-5.3-Flash now costs less.

Is GLM-4.6V open source?

Yes. GLM-4.6V's code is public under the MIT licence, so you can run it on your own server.

What are the best GLM-4.6V alternatives?

GLM-4.6V is an open source alternative to Hunyuan Vision 1.5 Thinking. Step3-VL-10B and Qwen3.6-27B do the same job and are open source too.

Hunyuan Vision 1.5 ThinkingPaid

Tencent's vision model for OCR, charts and visual reasoning on the TokenHub API.

Step3-VL-10BOpen Source

Choose it when you want a small open model for image reasoning and maths: StepFun's tests put it ahead of GLM-4.6V-Flash on MMMU (78.1 vs 71.2), though it reads images only and has no hosted API.

Qwen3.6-27BOpen Source

Choose it when you want to host a mid-size vision model yourself: it is a 27B dense model under Apache 2.0 that reads images and video with a 262,144-token context.

GLM-5.3-FlashOpen Source

Choose it when you call Z.ai's API: it is the newer multimodal model, cheaper at $0.15 / $0.50 per million tokens with a 1M-token context, though its 320B weights are far harder to host and its thinking cannot be turned off.

Questions about GLM-4.6V