GLM 5

Writing, coding and working through multi-step tasks as an agent are the jobs GLM 5 is built for. GLM 5 is an open-source large language model from Zhipu AI, a Tsinghua University spin-off, with 745 billion parameters and a focus on agentic intelligence and frontier-level reasoning.

It also covers writing and editing, which is why it sits under office and productivity.

The architecture is Mixture of Experts, with 44 billion parameters active at a time.

Only that active share runs for each request, which is how GLM 5 keeps costs lower than its size suggests.

Long-context processing uses DeepSeek's sparse attention mechanism.

The unusual detail is the hardware.

Zhipu AI trained GLM 5 entirely on Huawei Ascend chips with the MindSpore framework, with no dependence on other vendors.

The model is positioned as a cost-effective challenger to GPT-5 and Claude, and because it is open source you can inspect the weights instead of only calling an API.