Overview
GLM-5.2 is Z.ai's current flagship, the model the GLM-5 line has been building toward. It keeps the 744B-parameter mixture-of-experts design (40B active) of GLM-5 and DeepSeek Sparse Attention for efficient long-context inference, and pushes the context window to 1 million tokens — a direct answer to the coding-first use cases where GLM has made its name.
Z.ai positions GLM-5.2 as the open-weight model of choice for agentic engineering, and it performs strongly in coding harnesses like Claude Code, Cline and Roo Code. It's the latest proof that top-tier open-weight coding models can ship and compete globally, with self-hostable weights at low API cost.
Key capabilities
- 1-million-token context window
- 744B MoE with 40B active parameters
- Coding-first tuning for agentic engineering
- Open weights, low API pricing
At a glance
Params
744B total / 40B active
Context
1M tokens
Focus
Agentic coding
Pricing: Low open-weight pricing; self-hostable.
Pros & cons
What we like
- Frontier-class open-weight coding model
- Huge 1M context for whole-codebase work
- Excellent in popular coding harnesses
- Self-hostable under a permissive licence
Trade-offs
- Heavyweight to self-host at full size
- Closed frontier models still edge it on some tasks
The verdict
GLM-5.2 is the open-weight model to beat for agentic coding — big, capable and tuned for the long-horizon engineering work where it shines, now with a 1M context. For teams building on open models, it's one of the strongest foundations available today.
Best for: Agentic coding and long-horizon engineering on open weights.
Frequently asked questions
What's new in GLM-5.2?
It extends the GLM-5 architecture with a 1-million-token context window and further coding-first tuning, keeping the 744B MoE design and DeepSeek Sparse Attention for efficient inference.
Is GLM-5.2 open source?
Yes — GLM-5.2 ships with open weights, so you can self-host it subject to its licence terms.
