GLM-5.1

GLM-5.1 advances the GLM family’s focus on agentic engineering, improving coding and repository-oriented work over its predecessor. Its published evaluations report state-of-the-art results on SWE-Bench Pro and a substantial lead over GLM-5 on NL2Repo repository generation and Terminal-Bench 2.0 real-world terminal tasks. The model is also available through a publicly hosted open-weight repository and an NVIDIA NIM catalog entry, giving teams multiple deployment paths.

The model’s main strength is sustained work on complex software-engineering tasks rather than only producing an initial answer. It is described as breaking problems into steps, running experiments, inspecting results, recognizing blockers, and revising its strategy over hundreds of rounds and thousands of tool calls. This makes it a practical fit for engineering agents and long-running terminal workflows that benefit from iterative reasoning, while the linked technical report provides additional lineage and evaluation context.