GLM-5.1

754B parameter MoE built for long-horizon agentic engineering. Designed to sustain optimization over hundreds of rounds and thousands of tool calls without losing coherence. MIT License.

Achieves SOTA 58.4% on SWE-Bench Pro, plus 42.7% on NL2Repo, 63.5% on Terminal-Bench 2.0, 68.7% on CyberGym, and 68.0% on BrowseComp. Improved judgment on ambiguous problems and iterative reasoning revision compared to GLM-5. Available via SGLang, vLLM, xLLM, Transformers, and KTransformers.