Z.AI Updates & Release Notes

Native visual capabilities enable the model to observe interfaces, rendering results, and interaction feedback—creating a closed loop across code, browsers, and GUIs.

Efficient hybrid architecture: Combines linear and sparse attention with 320B total parameters and 18B activated, significantly reducing compute and KV-cache requirements.

Beyond coding: Supports office document and financial research workflows, autonomously breaking down goals, using tools, and refining outputs. Learn more in our documentation.*

Stronger Coding Capabilities

GLM-5.3 delivers a significant improvement in coding capabilities, achieving a 50% gain over GLM-5.2 on Z.ai Code Bench and reaching state-of-the-art (SOTA) performance among open-source models on public benchmarks, including Terminal Bench 3.0.

Emergent Cybersecurity Capabilities

GLM-5.3 matches Mythos 5 in white-box code review and vulnerability discovery. In collaboration with multiple cybersecurity teams, it has been tested on real-world targets and has identified a total of 2,436 vulnerabilities, including 1,097 medium- and high-severity vulnerabilities.Learn more in our documentation.*

Supports 1M lossless context, significantly improving long-horizon task capabilities and reducing context drift and goal forgetting in complex tasks

Achieves open-source SOTA performance on coding and long-horizon task benchmarks, delivering more stable results in complex system engineering and deep debugging

Significantly improves the real-world developer experience, with more reliable project-level context handling, adherence to engineering standards, and multi-platform development. Learn more in our documentation.*

Designed for long-horizon tasks, GLM-5.1 can work independently for up to 8 hours in a single run, enabling a full loop from planning and execution to iterative refinement and final delivery.

It demonstrates stronger engineering intelligence across autonomous planning, sustained execution, bug fixing, and strategy iteration, while achieving comprehensive capability alignment with Claude Opus 4.6. Built with multi-turn SFT, RL, and a process-quality evaluation framework, GLM-5.1 further improves stability, consistency, and tool use over extended tasks. Learn more in our documentation.*

Designed for vision-based coding and claw-style Agent workflows, GLM-5V-Turbo brings native multimodal understanding to images, video, and text, enabling stronger planning, coding, and action execution in visually grounded tasks.

It strengthens multimodal tool use and GUI task execution, allowing the model to better interpret interfaces, design mocks, documents, and charts, plan the next action, and complete end-to-end Agent workflows while maintaining strong pure-text coding and reasoning performance. Learn more in our documentation.*

We've launched GLM-OCR, a compact and high-performance optical character recognition model powered by the self-developed CogViT and GLM-0.5B encoder-decoder architecture, enabling efficient cross-modal alignment through its dedicated connection layer.

The update leverages CLIP pre-training on billions of image-text pairs to deliver robust visual semantic understanding and key token extraction capabilities, while maintaining a lightweight design for fast inference. Learn more in our documentation.*

GLM-4.7 Release Notes

We’ve released GLM-4.7, our latest flagship foundation model with significant improvements in coding, reasoning, and agentic capabilities. It delivers more reliable code generation, stronger long-context understanding, and improved end-to-end task execution across real-world development workflows.

The update brings open-source SOTA performance on major coding and reasoning benchmarks, enhanced agentic coding for goal-driven, multi-step tasks, and improved front-end and document generation quality. Learn more in our documentation.*