GLM-5.2 is probably the most powerful text-only open weights LLM (simonwillison.net)

Z.ai has recently launched GLM-5.2, touted as the most powerful text-only open weights large language model (LLM) to date. Released on June 16, 2026, with open weights under an MIT license, this model boasts an impressive 753 billion parameters and a 1 million token context window, significantly up from its predecessor, GLM-5.1, which supported only 200,000 tokens. With its Mixture of Experts architecture and cutting-edge performance, GLM-5.2 has quickly taken the lead on benchmarks such as the Artificial Analysis Intelligence Index, surpassing other notable models like MiniMax-M3 and DeepSeek V4 Pro. Despite its strengths, GLM-5.2 is characterized as token-hungry, utilizing 43,000 output tokens per task, higher than other models in its class. It also ranks second on the Code Arena WebDev leaderboard, indicating its strong capabilities in coding tasks despite lacking image input. This release is significant for the AI and machine learning community as it raises the bar for open-source LLMs, challenging the commercial status quo and offering developers a powerful tool for diverse applications at competitive pricing compared to renowned models like GPT-5.5 and Claude Opus 4.5-4.8.