Published: April 3, 2026
Z.ai (formerly Zhipu AI) has released GLM-5V-Turbo, a multimodal vision model with 744 billion parameters built on a Mixture-of-Experts architecture. The model combines native image and video understanding with code generation, targeting developers working in agentic engineering workflows.
Architecture and specifications
GLM-5V-Turbo uses a Mixture-of-Experts (MoE) design with 744 billion total parameters and 40 billion active per token. It was trained using what Z.ai calls 30+ Task Joint Reinforcement Learning, a methodology that optimizes the model across more than thirty distinct tasks simultaneously. This training approach aims to produce a model that can perceive visual inputs, plan actions, and execute code in a single pipeline.
The model supports a 202,752-token context window with a maximum output of 131,072 tokens. Pricing is set at $1.20 per million input tokens and $4.00 per million output tokens.
Benchmark performance
Z.ai reports that GLM-5V-Turbo scores 94.8 on the Design2Code benchmark, compared to 77.3 for Claude Opus 4.6. It also leads on GUI agent benchmarks including AndroidWorld and WebVoyager, suggesting strong performance for tasks involving browser and mobile interface interaction.
The model is positioned specifically for visual coding tasks: generating frontend code from design mockups, navigating graphical interfaces, and completing agentic browsing workflows. It is optimized for use with OpenClaw, the increasingly popular open-source coding assistant.
Company background
Zhipu AI, founded in 2019 as a spin-off from Tsinghua University, rebranded its international operations as Z.ai in July 2025 and went public on the Hong Kong Stock Exchange in January 2026. The company has been building its GLM model family for several years, with GLM-5V-Turbo representing its latest generation of multimodal AI.
Availability
GLM-5V-Turbo is available through Z.ai’s API platform. The model weights are also accessible through Hugging Face, making it available for developers who prefer to self-host.
For more information, visit Z.ai’s platform.
