The 750B parameter GLM-5.3 model surpasses Kimi K3 and rivals GPT-5.6-Sol on agentic coding benchmarks. Z.ai will release the open weights on Hugging Face in two weeks. This efficiency gain proves scaling post-training yields high performance with fewer parameters. Developers should monitor the API launch for frontier-level coding capabilities.