GLM-5.1

GLM-5.1 was introduced by Z.ai on April 7, 2026 as an open-weight flagship model built around sustained, long-horizon agentic execution rather than short interactive turns. According to the official release notes, the model is designed to work independently for up to eight hours in a single run, completing a full loop from planning and execution through iterative refinement to final delivery. It is positioned as a step toward broader engineering intelligence, with stronger capabilities in autonomous planning, sustained execution, bug fixing, and strategy iteration, and is described as achieving comprehensive capability alignment with Claude Opus 4.6 at release.

Beyond the headline long-horizon behavior, GLM-5.1 was trained with multi-turn supervised fine-tuning, reinforcement learning, and a process-quality evaluation framework aimed at improving stability, consistency, and tool use over extended tasks. Its direct successor, GLM-5.2, is explicitly framed on Z.ai's blog as a substantial leap in long-horizon task capability over GLM-5.1, which establishes 5.1 as the prior flagship in this agent-focused line. Practically, the model fits teams needing an open-weight coding and agentic assistant that can drive multi-hour engineering workflows on its own, while later variants in the family push further into million-token context and advanced agent coding.