Key takeaways

• GLM 4.6V is a large multimodal model designed for high-fidelity visual understanding and long-context reasoning across images, documents, and mixed media. It supports up to 128K tokens and enables interleaved image-text generation and UI reconstruction workflows.
• The model utilizes a Mixture of Experts (MoE) architecture and is open-weights under an MIT license, supporting native multimodal function calling.
• GLM 4.6V shows high proficiency in mathematical problem-solving, achieving an overall accuracy of 94.21% on the MATH-500 benchmark, excelling in breaking down multi-step problems and performing intricate calculations with detailed reasoning.
• It struggles significantly with complex or multi-step problem-solving tasks, specific tool usage, and complex code creation, as indicated by a low 17.5% accuracy on the Terminal-Bench (Terminus-2) evaluation.
• The model's limitations include occasional errors in complex arithmetic, misinterpretation of problem statements in geometric diagram analysis, and sensitivity to exact formatting in mathematical outputs.

Technical specifications

Context length

131k tokens

Max output

—

Architecture

MoE

Modalities

text+image->text

Parameters

~106B

Open weights

Yes

Licence

MIT

Regions

China

Vendor

z-ai

Provider model key

z-ai/glm-4.6v