GLM-5.1 hits 8-hour autonomous coding, tops SWE-Bench Pro

GLM-5.1 marks the transition from AI assistants to autonomous engineers capable of long-horizon project management. Its 8-hour autonomy window allows for thousands of tool calls and hundreds of iterations in a single session. Training on Huawei Ascend chips proves that state-of-the-art performance is achievable outside the Nvidia ecosystem. The combination of a 200,000-token context window and a 128,000-token output limit enables the generation and refactoring of massive codebases in one pass.