GLM-5.1 is Z.ai's spring 2026 iterative upgrade to GLM-5, open-sourced under MIT license on April 8, 2026. It retains the 744B/40B-active MoE architecture and Huawei Ascend training stack while improving reasoning, agentic workflows, and multilingual coverage.
GLM-5.1 is the spring 2026 iterative upgrade from [[China/Z.ai|Z.ai]] over its GLM-5 frontier-class model. Initially released to paid subscribers in late March 2026, GLM-5.1 was open-sourced on April 8, 2026 under MIT license, continuing Z.ai's strategy of treating open distribution as a competitive lever against U.S. closed-source frontier models. GLM-5.1 retains the GLM-5 MoE architecture (744B total / 40B active) and Huawei Ascend training stack while delivering improvements across reasoning, agentic workflows, and multilingual evaluations.
GLM-5.1 sits alongside [[DeepSeek/DeepSeek V4-Pro]], [[Alibaba Qwen/Qwen 3.5]], and [[Moonshot AI/Kimi K2.6]] as one of four Chinese open-weight frontier models released in the late-2025-through-spring-2026 window — collectively setting global price floors for capable open-weight LLMs.
Reasoning Improvements: Improved chain-of-thought stability and self-verification over GLM-5 baseline, particularly on multi-step reasoning chains.
Agentic Workflows: Continued leadership among open-source models on browser-driven and tool-using benchmarks (BrowseComp, MCP-Atlas) — incremental gains over GLM-5.
Multilingual: Strong Chinese and English performance with extended coverage across major non-English languages.
Open-Weight Distribution: Full weights publicly released under MIT, allowing commercial self-hosting.
As an iterative release, GLM-5.1 inherits GLM-5's hardware footprint (744B total parameters) and the same caveats about geopolitical procurement complications for some Western enterprise buyers. Some independent third-party benchmarks lag the model's release, so claimed deltas over GLM-5 are still being validated externally as of May 2026.
May 8, 2026