Zhipu AI Releases GLM-5.2, Open-Source Frontier Model
Key Questions
What are the key capabilities of Zhipu AI's GLM-5.2?
GLM-5.2 leads open-source models on Artificial Analysis with AIME 99.2, strong SWE-bench results ahead of GPT-5.5, and 1M context length under an MIT license.
How does GLM-5.2 achieve lower training costs?
It uses a critic (value model) for RL training on Huawei Ascend chips, resulting in 90% lower costs compared to prior approaches.
What trend does GLM-5.2 highlight in the AI model landscape?
The release underscores how open-weight models are becoming extremely cheap and competitive, challenging the viability of proprietary models.
GLM-5.2 (MIT license) leads open-source on Artificial Analysis (AIME 99.2, SWE-bench ahead of GPT-5.5, 1M context). Uses critic (value model) for RL training on Huawei Ascend chips at 90% lower cost. Highlights the trend of open-weight models becoming extremely cheap, challenging proprietary model viability.
Sources (2)
Updated Jun 25, 2026