GLM-5 is an open-weights artificial intelligence model specifically designed for complex systems engineering and long-horizon agentic tasks. Scaling from 355B parameters in GLM-4.5 to 744B parameters (40B active), the model increases pre-training data from 23T to 28.5T tokens while integrating DeepSeek Sparse Attention technology to reduce deployment costs while preserving long-context capacity.

The model features significant improvements in reasoning, coding, and agentic capabilities across academic benchmarks. It achieves best-in-class performance among all open-source models worldwide and demonstrates strong long-term planning and resource management abilities. GLM-5 can turn text or source materials directly into .docx, .pdf, and .xlsx files including PRDs, lesson plans, exams, spreadsheets, financial reports, and other ready-to-use documents.

GLM-5 utilizes DeepSeek Sparse Attention technology and incorporates slime, a novel asynchronous RL infrastructure that substantially improves training throughput and efficiency. This enables more fine-grained post-training iterations and bridges the gap between competence and excellence in pre-trained models. The model's architecture supports complex task completion through enhanced reasoning capabilities.

The model delivers significant improvements across frontend, backend, and long-horizon tasks compared to previous versions. On Vending Bench 2, which measures long-term operational capability over a one-year horizon simulating a vending machine business, GLM-5 ranks #1 among open-source models with a final account balance of $4,432, approaching Claude Opus 4.5 performance.

GLM-5 targets developers and researchers working with complex AI systems, available through multiple deployment options including coding agents like Claude Code and OpenCode, chat interfaces, and local deployment frameworks. It supports various hardware platforms including non-NVIDIA chips through kernel optimization and model quantization.