GLM-5: Zhipu AI's massive new language model takes on GPT-4
Zhipu AI just rolled out GLM-5, a massive open-source language model with 745 billion parameters.
It's built to handle everything from coding and creative writing to long conversations and complex reasoning, making it a serious player in the AI scene.
GLM-5 uses DeepSeek sparse attention tech
GLM-5 uses smart tech like DeepSeek Sparse Attention to process over 128,000 tokens efficiently.
It was trained entirely on Huawei chips (no US hardware involved) using the MindSpore framework—pretty cool for tech independence.
This version is a big step up from previous GLM-4.x models.
GLM-5 goes toe-to-toe with top models
On coding tasks, GLM-5 goes toe-to-toe with top models like Anthropic's Claude Opus. You can already try it through an API, and open-source weights are available as of February 2026.
Zhipu AI is now listed in Hong Kong and is looking beyond China for new users.