GLM-5.2: Z.ai Releases Most Powerful Text-Only Open Weights LLM
Key Takeaways
- ▸GLM-5.2 is now the highest-ranked open weights LLM on the Artificial Analysis Intelligence Index with a score of 51
- ▸The model supports a 1-million token context window and operates at a competitive price point of ~$1.40/$4.40 per million tokens
- ▸Despite lacking multimodal vision capabilities, GLM-5.2 ranks 2nd on Code Arena WebDev leaderboard for front-end development tasks
Summary
Chinese AI lab Z.ai released GLM-5.2, a 753-billion parameter text-only language model with open weights under an MIT license on June 16th, 2026. The model features 40 active parameters through a Mixture of Experts architecture and supports a 1-million token context window—a significant increase from GLM-5.1's 200,000 tokens. GLM-5.2 represents a major milestone for open-source AI development, offering competitive performance to closed-source alternatives at substantially lower costs.
According to Artificial Analysis' widely respected Intelligence Index, GLM-5.2 ranks as the leading open weights model with a score of 51, surpassing MiniMax-M3 (44) and DeepSeek V4 Pro (44). The model also achieved 2nd place on the Code Arena WebDev leaderboard for front-end development tasks, behind only Claude Fable 5, demonstrating strong performance in agentic coding workflows despite lacking multimodal vision capabilities. However, the model exhibits higher token consumption per task than competitors, using 43,000 output tokens per Intelligence Index task compared to 26,000 for GLM-5.1.
Available through multiple providers via OpenRouter at approximately $1.40 per million input tokens and $4.40 per million output tokens, GLM-5.2 offers compelling economics compared to closed alternatives like GPT-5.5 ($5/$30) and Claude Opus 4.5-4.8 ($5/$25). The open availability of such a powerful model democratizes access to cutting-edge LLM capabilities for researchers, developers, and organizations.
- The 753B parameter model trades token efficiency for performance, consuming more output tokens per task than competitors
