GLM-5.1

GLM-5.1 is a free open-source large language model from Z.ai (zai-org) hosted on Hugging Face, designed for complex systems engineering, long-horizon agentic tasks, reasoning, and tool-calling workflows. The model is distributed at no cost under an open-weights license, making it suitable for researchers, AI engineers, and enterprises seeking frontier-grade open models.

The GLM-5 architecture scales to 744B total parameters with 40B active parameters per forward pass (a Mixture-of-Experts design), up from GLM-4.5's 355B/32B configuration. Pre-training data was expanded from 23T to 28.5T tokens, and the model integrates DeepSeek Sparse Attention (DSA) to substantially reduce deployment cost while preserving long-context capacity. Z.ai's team also developed 'slime,' a novel asynchronous reinforcement learning infrastructure that improves RL training throughput, enabling fine-grained post-training iterations. On benchmarks, GLM-5 scores 30.5 on Humanity's Last Exam (50.4 with tools), 92.7 on AIME 2026 I, 96.9 on HMMT Nov. 2025, 86.0 on GPQA-Diamond, 77.8 on SWE-bench Verified, and 73.3 on SWE-bench Multilingual — closing the gap with frontier closed models like Claude Opus 4.5, Gemini 3 Pro, and GPT-5.2.

Developer deployment is flexible: the model runs through Hugging Face Transformers, vLLM, SGLang, Docker Model Runner, llama.cpp, Ollama, and LM Studio. It exposes an OpenAI-compatible chat completions API when self-hosted, and is also offered as a managed service via the Z.ai API Platform. Based on our analysis of 870+ AI tools, GLM-5.1 stands out among open-weights LLMs by combining best-in-class open-source performance on reasoning, coding, and agentic benchmarks with multiple production-ready inference paths — giving teams an alternative to API-only frontier models when they need data sovereignty, custom fine-tuning, or on-prem control.