GLM-4.6V

GLM-4.6V (December 8, 2025) is Zhipu's open-weight multimodal flagship: a 106B model pairing vision with the GLM-4.6 language backbone, plus a 9B GLM-4.6V-Flash variant tuned for local deployment. Its signature feature is native function calling where images, screenshots and document pages pass directly as tool parameters — built for the agentic workflows Zhipu targets. Like the rest of the post-4.5 GLM line, it ships under the MIT license with weights on Hugging Face and ModelScope, runnable via vLLM and SGLang.

Context: this extends the GLM-4.6 family (the 357B MoE coding model already in the catalog) into vision. The 106B size means serious GPU requirements for the full model — the 9B Flash is the practical self-host option. Zhipu's open-everything cadence since GLM-4.5 (4.6, 4.7, GLM-5, 5.1, 5.2) is the most aggressive open-weight program among frontier labs, though some commercial API tiers stay closed.