Z.ai exec tests new GLM-5.1 against company's own services

Headline

Z.ai’s Product Director Tests Whether New GLM-5.1 Model Outperforms Company’s Own Services

Summary

Zixuan Li, Z.ai’s Director of Product and genAI Strategy, tweeted about switching to the company’s newly released GLM-5.1 model to see if it overwhelms their existing services. GLM-5.1 launched on March 27, 2026. It’s a refined version of their 744B-parameter Mixture of Experts model with DeepSeek Sparse Attention. The model scored 77.8% on SWE-bench Verified, putting it close to proprietary models like Claude Opus 4.5. Z.ai offers coding plans starting at $10/month.

Analysis

Li previously worked as an AI researcher at MIT before joining Z.ai. His tweet builds hype around GLM-5.1’s lower deployment costs and strong performance on complex engineering tasks. By framing this as a test that might “crush” their own infrastructure, he’s positioning GLM-5.1 as a serious upgrade for AI coding agents. The model works with tools like Claude Code and Cursor, which could pull users away from OpenAI and Anthropic.

Z.ai was founded in 2019 and went public in 2026. Like other Chinese AI companies, they’ve open-sourced their models to build developer communities outside China. The promotional tone here suggests this is marketing as much as technical commentary—possibly a response to user complaints about compute availability.

Impact Assessment

• Significance: Medium
• Categories: Model Release, Open Source, Developer Tools