Zhipu AI launches open-source GLM-5.1 model for coding tasks

Z.ai has rolled out GLM-5.1, a new flagship model built for agentic engineering and long-horizon coding work. The company is pitching it as a model that moves past one-shot code generation and can stay on a single job for up to eight hours, handling planning, execution, testing, fixes, and repeated optimization before returning production-ready output. Z.ai says this is one of the few models it has evaluated at that duration, and the first Chinese model in its testing to reach that level.

What stands out is the jump in coding and autonomous execution. Z.ai’s documentation says GLM-5.1 scores 58.4 on SWE-Bench Pro, ahead of GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro, while the GitHub materials say it also opens a wider gap over GLM-5 on repo generation and real terminal work. The technical profile is aimed at developers rather than casual chat users, with a 200K context window, 128K maximum output, optional deep thinking, and streamed tool-call output for live agent workflows.

This launch is not limited to a benchmark post. Z.ai says GLM-5.1 is already available to all GLM Coding Plan users across Max, Pro, and Lite, with setup guides for Claude Code, OpenClaw, Cline, and other OpenAI-compatible coding tools. The company has also published downloadable GLM-5.1 and GLM-5.1-FP8 weights in its open model repo, while noting that chat.z.ai support is still on the way.

The release also says a lot about where the company is heading. Z.ai, known previously as Zhipu AI outside China, has been pushing the GLM line deeper into coding and agent infrastructure, including OpenClaw-oriented models and broader tool integrations. Reuters previously described the company as one of China’s AI tigers and noted its Hong Kong listing earlier this year.

Source