Overview
GLM-5 is Z.ai's flagship and, on launch, the leading open-weights model on the Artificial Analysis Intelligence Index. It marked the first new GLM architecture since 4.5, scaling from 355B to 744B total parameters (40B active) and 28.5T training tokens, while integrating DeepSeek Sparse Attention to keep deployment cost and long-context performance in check.
Z.ai positions GLM-5 for complex systems engineering and long-horizon agentic tasks — exactly the work where it makes large gains over GLM-4.7 on economically valuable benchmarks. It's a favourite in coding harnesses like Claude Code, Cline and Roo Code, and the follow-up GLM-5.1 and GLM-5.2 releases have pushed coding scores and context length (up to 1M tokens) even further.
Key capabilities
- Top open-weights model on intelligence index at launch
- 744B MoE with DeepSeek Sparse Attention
- Built for agentic engineering and long-horizon tasks
- Strong in popular coding harnesses
At a glance
Params
744B total / 40B active
Open weights
Leading index rank
Focus
Agentic engineering
Pricing: Low open-weight pricing; self-hostable.
Pros & cons
What we like
- Best-in-class open-weight intelligence at launch
- Excellent for agentic coding workflows
- Strong long-context and reasoning
Trade-offs
- Heavyweight to self-host at full size
- Closed frontier models still lead on some tasks
The verdict
GLM-5 is the open-weight model to beat for agentic engineering — big, capable and tuned for the long-horizon coding work where it shines. For teams building on open models, it's one of the strongest foundations available.
Best for: Agentic coding and long-horizon engineering on open weights.
Frequently asked questions
What makes GLM-5 good at coding?
It's tuned for agentic engineering and long-horizon tasks, performs strongly in harnesses like Claude Code and Cline, and the 5.1/5.2 updates pushed its coding benchmark scores to the top of the open-weight field.
Is GLM-5 open source?
Yes — GLM-5 ships with open weights, so you can self-host it, subject to its licence terms.
