GLM-5.2: Built for Long-Horizon Tasks

GLM-5.2 is a new flagship model designed for long-horizon tasks, featuring a reliable 1M-token context and advanced coding capabilities. It introduces an architectural improvement called IndexShare, which reuses an indexer across multiple sparse attention layers to significantly reduce computational costs at long context lengths. The model demonstrates competitive performance on several long-horizon coding benchmarks, such as FrontierSWE and PostTrainBench, ranking as a top open-source model. GLM-5.2 also offers flexible effort levels to balance performance and latency and is released under an open-source MIT license.