GLM-5-Turbo

GLM-5-Turbo is Z.ai's agent-focused variant, released March 15, 2026 and tuned from the ground up for OpenClaw and other complex agentic workflows rather than retrofitted from a general-purpose base. It builds on the GLM-5 architecture (744 billion parameters, around 40 billion active), supports a 200,000-token context window with up to 128,000-token output, and offers selectable thinking modes that trade compute for reasoning depth. What sets GLM-5-Turbo apart is its alignment for production agentic tasks: stable, accurate tool invocation, decomposition of multi-layered instructions into sub-steps, awareness of scheduled and persistent long-running tasks, and faster, more stable execution in high-throughput, long-chain workflows common in business automation. It can take tool descriptions, skill references, workspace details, and session state directly in its system prompt for modular instruction handling. Key capabilities include streaming output, function calling, context caching for extended conversations, structured JSON output, and Model Context Protocol (MCP) tool integration. Compared with models adapted for agents after training, GLM-5-Turbo inherits GLM-5's long-horizon planning and agentic coding strengths while running faster, making multi-step autonomous workflows more practical at production scale. Thinking mode enables internal reasoning before a response, with lighter modes for extraction and classification and deeper modes for complex instruction following and logical decomposition.