China’s Z.ai Open-sources GLM-5.1, a Long‑running AI Agent Built for 8‑hour Tasks

Chinese AI startup Z.ai, also known as Zhupai AI, has released a new flagship large language model, GLM-5.1, as open source under the permissive MIT license positioning it as an AI “marathon runner” designed to work autonomously on a single task for up to eight hours.

The model is available for download and commercial use on Hugging Face, where enterprises can pull it into their own infrastructure, customize it and deploy it without restrictive licensing terms.

Z.ai is best known for its GLM family of models. Last month, the company introduced GLM-5 Turbo, described as a faster variant, but offered that model only under a proprietary license. GLM-5.1, by contrast, arrives squarely in the open-source ecosystem, with the MIT license enabling broad reuse.

Where many AI developers have been focused on “reasoning tokens” larger context windows and more capable step-by-step reasoning — Z.ai is optimizing for what it calls “productive horizons”: how long an AI agent can stay on track, aligned with a goal, over extended periods of autonomous work.

GLM-5.1 is a 754-billion-parameter Mixture-of-Experts (MoE) model. According to Z.ai, it is engineered to maintain goal alignment over long execution traces that can span thousands of tool calls, a setup aimed at enabling agents that can handle complex, multi-step workflows rather than short bursts of assistance.

The company frames this as a shift “from vibe coding to agentic engineering”  moving away from casual, one-off code suggestions toward more structured, end-to-end task execution that can run for hours without constant human intervention.

Eight-hour agents and a new open-source benchmark

In a post on X, Z.ai leader Lou contrasted the current generation of agents with GLM-5.1’s extended capabilities, writing that “agents could do about 20 steps by the end of last year” and that “glm-5.1 can do 1,700 rn.” Lou went on to argue that “autonomous work time may be the most important curve after scaling laws,” and described GLM-5.1 as “the first point on that curve that the open-source community can verify with their own hands,” before adding, “hope y’all like it^^”.

The model is designed to work autonomously on a single task for up to eight hours, underscoring Z.ai’s focus on long-running AI agents rather than just faster response times. In an AI market “increasingly crowded with fast models,” the company characterizes GLM-5.1 as the long-distance runner among LLMs.

The open-source release also arrives amid renewed questions about China’s role in the open AI ecosystem. GLM-5.1, as part of a well-known Chinese-built model family and released under a permissive license, could become a widely examined reference point for long-horizon agent behaviour.

Z.ai, which is listed on the Hong Kong Stock Exchange and reported a market capitalisation of $52.83 billion at the time of its listing in early 2026, is using GLM-5.1 to further cement its position in the global AI landscape.