GLM-5.1 is a next-generation flagship model designed specifically for agentic engineering, representing a major step beyond its predecessor in autonomous task execution. With a reported 754 billion parameters, the model is built to handle complex, long-horizon software engineering work rather than short, transactional responses. Its architecture is tuned for sustained optimization, allowing it to break down complicated problems into stages, run experiments, read outputs, and course-correct through repeated iteration. Rather than exhausting its repertoire early and plateauing—as previous models including GLM-5 tend to do—GLM-5.1 maintains productive momentum over extended sessions, sustaining hundreds of optimization rounds and thousands of tool calls. This endurance makes it particularly suited for tasks requiring multi-hour autonomous execution.

The model achieves state-of-the-art results on SWE-Bench Pro, outperforming its own family line and rivaling top closed-source models on complex software engineering benchmarks. It leads GLM-5 by a substantial margin on NL2Repo repository generation and Terminal-Bench 2.0 real-world terminal tasks, demonstrating that its capabilities extend beyond synthetic benchmarks into practical engineering scenarios. Zhipu AI has released GLM-5.1 with open weights on HuggingFace, a technical report on arxiv, and full documentation on GitHub, alongside API availability. The company has pursued aggressive price increases this year, signaling a push to monetize advanced capabilities while keeping costs competitive with U.S. peers. GLM-5.1 is positioned as a practical tool for teams and organizations needing autonomous coding agents that can work reliably over hours rather than minutes.