GLM-5V-Turbo

// Overview

GLM-5V-Turbo is the multimodal vision model in Z.ai (Zhipu / GLM)'s GLM Turbo line, released 1 April 2026 and now the current flagship of that line. Unlike a text model with a bolt-on image reader, GLM-5V-Turbo treats visual perception as a core part of reasoning, planning and execution: it takes images, short video clips, design drafts and document layouts directly in context and produces code, structured actions or GUI steps as output.

The headline use case is vision-to-code. GLM-5V-Turbo is tuned to turn a design mockup, screenshot or product walkthrough into runnable front-end code, and it is explicitly integrated for agentic coding workflows such as OpenClaw and Claude Code. Architecturally it builds on the GLM-5-Turbo text base, adding the CogViT vision encoder and a Multimodal Multi-Token Prediction (MMTP) path so it keeps competitive text-only coding ability while gaining native multimodal grounding.

GLM-5V-Turbo is served only through Z.ai's API and chat platform — unlike the open-weights GLM-5 flagship, this Turbo vision variant ships proprietary with no published weights. It exposes a 200K-token context window and up to 128K tokens of output, supports thinking mode, streaming, function calling and context caching, and is priced at $1.20 per million input tokens and $4.00 per million output tokens.

// Benchmarks

This model's scores

1. Design2Code (multimodal coding)94.8%
2. AndroidWorld (GUI agents)75.7%
3. OSWorld (GUI agents)62.3%
4. MMSearch (multimodal tool use)72.9%
5. ImageMining (visual search)30.7%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Same per-token pricing as the GLM-5-Turbo text model. API-only via the Z.ai platform.

Pricing source ↗

// Strengths

• Native multimodal grounding — images, video, design drafts and document layouts are processed as first-class reasoning inputs, not OCR'd afterthoughts.
• Vision-to-code: converts mockups, screenshots and walkthroughs into executable front-end code, the model's primary design goal.
• Built for agentic engineering — explicitly integrated with OpenClaw and Claude Code workflows, with function calling and context caching.
• Keeps competitive text-only coding ability via the GLM-5-Turbo base while adding multimodal perception.
• Large 200K context with up to 128K output, suited to long, tool-augmented multi-step tasks.

// Best for

• Turning Figma/design mockups and UI screenshots directly into working front-end code.
• GUI and computer-use agents that read a screen and take multi-step actions (Android, desktop, web).
• Multimodal document understanding — extracting structure and answers from complex PDF and document layouts.
• Visual tool use and multimodal search inside agent frameworks.
• Video-grounded tasks such as reasoning over screen recordings and product walkthroughs.

// How to access

// GLM Turbo — every version

The full lineage of the GLM Turbo line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.