Z.AI's GLM-5.1 Achieves SOTA on SWE-Bench Pro with 58.4% Score, Beating GPT-5.4 and Claude Opus

Chinese AI startup Z.AI released GLM-5.1, an open-source 754B parameter model with MIT licensing that scored 58.4% on SWE-Bench Pro, surpassing GPT-5.4 (57.7%) and Claude Opus 4.6 (57.3%). The model features 'agentic engineering' design for autonomous 8-hour task execution, mixture-of-experts architecture, and 202,752 token context window with innovative 'stepped optimization' mechanism to avoid performance plateaus. GLM-5.1 represents a significant advance in open-source coding capabilities, challenging closed-source model dominance in software engineering tasks.