zai-org/GLM-5

GLM-5: From Vibe Coding to Agentic Engineering

What it solves

GLM-5 is a series of flagship large language models designed to handle complex systems engineering and long-horizon agentic tasks. It addresses the limitations of previous models that often plateau during complex problem-solving, providing the ability to maintain productivity over hundreds of rounds of iteration and thousands of tool calls.

How it works

The series evolves through three main versions:

  • GLM-5: Scales to 744B parameters (40B active) and utilizes DeepSeek Sparse Attention (DSA) to reduce deployment costs while maintaining long-context capacity. It was trained using "slime," an asynchronous RL infrastructure for efficient post-training.
  • GLM-5.1: Focuses on agentic engineering and coding, improving the model's ability to break down complex problems, run experiments, and revise strategies over long sessions.
  • GLM-5.2: Introduces a solid 1M-token context window and the "IndexShare" architecture, which reuses indexers across sparse attention layers to reduce per-token FLOPs by 2.9×. It also features an improved MTP layer for speculative decoding.

Who it’s for

Developers and researchers working on autonomous agents, complex software engineering, and tasks requiring massive context windows (up to 1 million tokens) or high-precision reasoning and coding.

Highlights

  • Massive Context: GLM-5.2 supports a stable 1M-token context window.
  • Agentic Performance: Specifically optimized for long-horizon tasks, ranking #1 among open-source models on Vending Bench 2.
  • Flexible Reasoning: Supports a reasoning_effort parameter (max or high) to balance performance and latency.
  • High Efficiency: Uses IndexShare and DSA to optimize computation and deployment costs.