GLM-5.1

GLM-5.1 is positioned by Z.ai as a next-generation flagship aimed squarely at agentic engineering, with the central design goal of keeping software-build workflows productive across much longer sessions rather than plateauing early. The model introduces architectural and training changes that the team credits with sustained effectiveness on ambiguous problems, iterative experimentation, and multi-step debugging. Compared with its predecessor, GLM-5.1 shows substantially stronger coding ability while preserving a broad agentic skill set.

On quantitative grounding, GLM-5.1 reports a state-of-the-art 58.4 score on SWE-Bench Pro, ahead of the listed comparators that include GLM-5, GPT-5.4, Opus 4.6, and Gemini 3.1 Pro. It also widens the margin over GLM-5 on NL2Repo repository generation and Terminal-Bench 2.0, signaling real-world terminal and end-to-end repo-building strength. Practical fit centers on long-running coding agents, repository synthesis, and cybersecurity-adjacent engineering tasks where the model can iterate, revisit reasoning, and break complex goals into smaller experiments with consistent follow-through.