GLM 4.6V

GLM 4.6V - Native Multimodal Function Calling Vision-Language Model

GLM 4.6V is Z.AI's (Zhipu AI) open-source vision-language model that introduces native multimodal function calling - bridging the gap between visual perception and executable action to provide a unified foundation for multimodal agents.

Native Multimodal Function Calling

GLM 4.6V's defining capability is its closed-loop perception-to-execution pipeline:

• Multimodal Tool Input: Images, screenshots, and document pages are passed directly as tool call parameters - no prior text conversion required, preventing information loss.
• Multimodal Tool Output: Visual results returned by tools (charts, rendered web screenshots, search-retrieved product images) are directly interpreted and incorporated into subsequent reasoning chains without text intermediaries.

Documented Capabilities

• Interleaved Image-Text Generation: Takes multimodal context (documents, user inputs, tool-retrieved images) and synthesizes coherent, visually grounded mixed-media outputs. During generation, the model can autonomously call search and retrieval tools to curate additional visuals.
• Multimodal Document Understanding: Processes up to 128K tokens of multi-document input as direct image pages - interpreting text, layout, charts, tables, and figures jointly without OCR preprocessing. Equivalent to ~150 complex document pages, 200 slide pages, or a 1-hour video in a single inference pass.
• Frontend Replication & Visual Editing: Reconstructs pixel-accurate HTML/CSS/JS from UI screenshots by detecting layout, components, and color schemes. Supports natural-language-driven iterative visual edits - users can circle a region and issue a text instruction; the model locates and corrects the corresponding code.