Z.ai has launched GLM-5, a massive 744B-parameter model designed for complex systems engineering and long-horizon agentic tasks. The model scales from GLM-4.5's 355B parameters to 744B (40B active) and expands training data from 23T to 28.5T tokens, while integrating DeepSeek Sparse Attention to reduce deployment costs.

GLM-5 represents a significant leap in reinforcement learning infrastructure through slime, an asynchronous RL system that improves training throughput and enables fine-grained post-training iterations. The model achieves best-in-class performance among open-source models on reasoning, coding, and agentic tasks, closing the gap with frontier models like Claude Opus 4.5. On the Vending Bench 2 benchmark, GLM-5 managed a simulated vending machine business to $4,432 over a one-year horizon.

Available under the MIT License on Hugging Face and ModelScope, GLM-5 is accessible through Z.ai's platform with built-in skills for document creation. The model can generate ready-to-use .docx, .pdf, and .xlsx files for PRDs, financial reports, and lesson plans. With benchmark scores showing 77.8% on SWE-bench Verified and 56.2% on Terminal-Bench 2.0, GLM-5 positions itself as a powerful tool for developers transitioning from vibe coding to agentic engineering.