GLM-5: Advanced Agentic AI for Complex Tasks

GLM-5, developed by zai-org, represents a significant advancement in large language models, specifically targeting complex systems engineering and long-horizon agentic tasks. This model scales substantially from its predecessor, GLM-4.5, featuring 744 billion parameters (40B active) and trained on an extensive 28.5 trillion tokens.

Key Innovations & Capabilities

• Efficient Long-Context Processing: Integrates DeepSeek Sparse Attention (DSA) to reduce deployment costs while maintaining robust long-context capabilities.
• Enhanced Reinforcement Learning: Utilizes 'slime', a novel asynchronous RL infrastructure, to improve training throughput and efficiency for fine-grained post-training iterations.
• Superior Performance: Achieves best-in-class results across various academic benchmarks for reasoning, coding, and agentic tasks among open-source models, closing the gap with frontier models.

Performance Highlights

GLM-5 demonstrates strong performance across a diverse set of benchmarks, including:

• HLE (Humanity's Last Exam): Scores 30.5 (text-only) and 50.4 (with tools), indicating advanced reasoning and tool-use capabilities.
• SWE-bench Verified: Achieves 77.8, showcasing its proficiency in code generation and problem-solving.
• Terminal-Bench 2.0: Scores 56.2 / 60.7 on Terminus 2 and 56.2 / 61.1 on Claude Code, highlighting its strength in terminal-based tasks.
• BrowseComp: Reaches 62.0 (without context management) and 75.9 (with context management), demonstrating strong browsing and information retrieval skills.

Deployment

GLM-5 supports local deployment via vLLM, SGLang, KTransformers, and xLLM, with specific Docker images and installation instructions provided for various GPU architectures and NPUs.