Overview
GLM-5V-Turbo is Z.ai's first native multimodal vision coding model, designed to bridge the gap between visual perception and code execution. Released on April 1, 2026, it processes images, videos, design drafts, and document layouts as primary inputs while generating code and executing multi-step agent workflows.
Unlike GLM-5.1 which focuses on text-based coding performance, GLM-5V-Turbo adds a CogViT vision encoder and 30+ task joint reinforcement learning pipeline to handle vision-grounded tasks. It targets developers building GUI automation agents, design-to-code pipelines, and document processing workflows where understanding visual context is essential for accurate code generation.
What's New
Native Multimodal Vision Coding
GLM-5V-Turbo natively fuses visual understanding with code generation from pretraining through post-training. The model processes images, videos, design drafts, and complex document layouts as first-class inputs, reducing the need for separate vision preprocessing in many multimodal coding workflows. This enables direct design-to-code generation, resume screening, visual grounding, and document-grounded writing in a single model call.
CogViT Vision Encoder
GLM-5V-Turbo introduces a new CogViT vision encoder combined with an inference-friendly architecture optimized for coding-relevant visual tasks. It handles high-resolution design mockups, screenshots of GUI environments, and multi-page documents while maintaining the 200K context window and 128K max output token capacity needed for long-horizon code generation.
Agent Workflow Optimization
GLM-5V-Turbo completes the full perceive-plan-execute loop for autonomous environment interaction. It delivered strong performance on AndroidWorld and WebVoyager for GUI agent tasks, solid results across CC-Bench-V2 benchmarks, and competitive scores on PinchBench, ClawEval, and ZClawBench for agent task execution. The model integrates natively with OpenClaw and Claude Code workflows.
30+ Task Joint Reinforcement Learning
The model was trained with joint reinforcement learning across STEM reasoning, visual grounding, video understanding, and coding domains simultaneously. This multi-task approach enables GLM-5V-Turbo to handle diverse visual inputs — from mathematical diagrams to UI wireframes — while maintaining strong code generation capabilities across all input types.
Performance Benchmarks
GLM-5V-Turbo scores 43 on the Artificial Analysis Intelligence Index, according to Artificial Analysis.
| Benchmark | Category | GLM-5V-Turbo |
|---|---|---|
| AndroidWorld | GUI agent tasks | Strong |
| WebVoyager | Web navigation agent | Strong |
| CC-Bench-V2 | Backend + frontend coding | Strong |
| ZClawBench | Agent task execution | Strong |
| PinchBench | Agent evaluation | Competitive |
| ClawEval | Agent evaluation | Competitive |
The model supports text, image, and video input with a 200K token context window. Performance data is based on Z.ai's official benchmark reports.
Pricing & Plans
GLM-5V-Turbo is available through Z.ai's API with pay-per-use pricing.
API Pricing
- Input tokens: $1.20 per 1M tokens
- Output tokens: $4.00 per 1M tokens
- Context window: 200K tokens
- Max output: 128K tokens
Comparison with Other Z.ai Vision Models
| Model | Input Cost (1M tokens) | Output Cost (1M tokens) | Status |
|---|---|---|---|
| GLM-5V-Turbo | $1.20 | $4.00 | Current |
| GLM-4.6V-Flash | Free | Free | Available |
| GLM-OCR | $0.03 | — | Available |
For chat-based access, GLM-5V-Turbo is available through the Z.ai platform. Subscription plan access details may vary — check the official pricing page for current availability.
Best For
- Frontend developers converting design mockups and Figma exports into production code
- QA engineers building GUI automation agents for Android and web application testing
- Document processing teams extracting structured data from PDFs, resumes, and scanned layouts
- Agent developers building perceive-plan-execute workflows with OpenClaw or Claude Code
- Teams needing vision-grounded code generation for screenshots, diagrams, and UI wireframes
FAQ
How does GLM-5V-Turbo differ from GLM-5.1?
GLM-5.1 is a text-focused coding model optimized for IDE integration and agentic coding tasks. GLM-5V-Turbo adds native multimodal vision capabilities through the CogViT encoder, enabling it to process images, videos, and design drafts as primary inputs for code generation. Choose GLM-5.1 for pure text coding tasks and GLM-5V-Turbo when visual context matters.
What input types does GLM-5V-Turbo support?
The model accepts images (PNG, JPG, WebP), video files, PDF documents, Word documents, and text. It can process design mockups, screenshots, diagrams, and multi-page document layouts alongside text prompts.
Can I use GLM-5V-Turbo with existing coding tools?
Yes. GLM-5V-Turbo integrates with OpenClaw and Claude Code workflows. It supports function calling for external tool integration, streaming output for real-time responses, and context caching for extended conversations.
Is GLM-5V-Turbo available for self-hosting?
Check Z.ai's official documentation for the current deployment and availability options for GLM-5V-Turbo weights and self-hosted deployment.
How does the pricing compare to other multimodal models?
At $1.20 per 1M input tokens and $4.00 per 1M output tokens, GLM-5V-Turbo is positioned as a cost-effective option for vision coding tasks. Z.ai also offers GLM-4.6V-Flash as a free alternative for lighter vision workloads.
