Zhipu launches GLM-5.1 for coding agents

Hot take: this is less a routine model refresh and more a direct attempt to define the long-running coding agent category.

• –The 8-hour autonomous execution claim is the clearest differentiator and speaks to real engineering workflows rather than just benchmark optics.
• –Zhipu is positioning GLM-5.1 as a systems model for planning, execution, fixing, and iterating, not just code generation.
• –The official docs explicitly compare it with Claude Opus 4.6, which signals an aggressive top-tier competitive push.
• –The release also emphasizes process quality, multi-turn SFT, and RL, suggesting stability over long task chains is a key product goal.