GLM-5.1 is a 744B-parameter mixture-of-experts model with 40B active parameters, supporting a 200K-token context and released under the MIT license. It achieves top performance on the SWE-Bench Pro benchmark.