Li Zixuan, Global Head of Z.ai (formerly Zhipu AI), posted on X Thursday evening: "Don't panic. GLM-5.1 will be open source." That was the entire message. No release date, no benchmarks, no specs.
The timing matters. Z.ai shipped GLM-5-Turbo on March 16, a closed-source, agent-focused variant of the flagship GLM-5. The company explicitly said it had no commitment to open-sourcing the Turbo model, prompting speculation that Z.ai was drifting away from the MIT-licensed approach that made GLM-5 popular. Li's post reads like a direct response to that anxiety.
GLM-5, released on Hugging Face in February under MIT, is a 744-billion-parameter mixture-of-experts model with 40 billion active parameters. It quickly topped open-source leaderboards in coding and agentic benchmarks, trained entirely on Huawei Ascend chips with no NVIDIA dependency. Z.ai priced the API at roughly $1 per million input tokens, undercutting Western frontier models by a wide margin.
GLM-5-Turbo narrowed that open ethos. Built specifically for Z.ai's OpenClaw agent ecosystem, it offered lower tool-call error rates and faster throughput, but only via API. The pivot drew attention given broader tensions in China's open-source AI scene: Alibaba's Qwen division recently saw its third senior executive departure of 2026, and CEO Eddie Wu took direct control of a new AI business group amid questions about open-model profitability.
What GLM-5.1 actually brings, and when, remains unknown. Li's four-word reassurance is thin on substance but clear on intent: Z.ai's next flagship stays open.
Bottom Line
Z.ai's next model iteration, GLM-5.1, will be released as open source, though no timeline or technical details have been shared.
Quick Facts
- Li Zixuan posted the announcement on X on March 20, 2026
- GLM-5-Turbo (closed-source) launched March 15-16, 2026
- GLM-5 is 744B parameters, 40B active, MIT-licensed
- Z.ai IPO'd in Hong Kong on January 8, 2026, raising ~$558M
- No release date or specs disclosed for GLM-5.1
