ZAI released GLM-4.7 Flash, a lightweight 30B parameter MoE model with 128K token context window, optimized for coding, DevOps, and aesthetics. Outperforms competitors in SWE-bench (59.2%) and AIME 25 (91.6%). Open source under MIT license. Suitable for local deployment with GPUs or quantized versions. Huggingface ZAI org