Z.ai released GLM-5.2 under an MIT license on June 16th. The model is 753 billion parameters, 1.51TB on disk, uses a Mixture of Experts architecture with 40 billion active parameters, and supports a 1 million token context window, up from 200,000 in GLM-5.1. It is text-only. According to Artificial Analysis, it now leads all open weights models on the Intelligence Index v4.1 with a score of 51, ahead of MiniMax-M3 at 44, DeepSeek V4 Pro at 44, and Kimi K2.6 at 43. It also sits second on Code Arena's WebDev leaderboard, behind only Claude Fable 5, despite having no vision input.
The tradeoff is token consumption. GLM-5.2 burns 43,000 output tokens per Intelligence Index task, compared to 26,000 for GLM-5.1 and 37,000 for DeepSeek V4 Pro. That matters for cost. Nine providers on OpenRouter currently price it at $1.40 per million input tokens and $4.40 per million output tokens, well below GPT-5.5 at $5 and $30, and Claude Opus 4.5 at $5 and $25. At that token burn rate, the gap narrows fast.
The original post is worth reading for the SVG generation tests alone. The author runs a consistent benchmark using animated SVG prompts, and GLM-5.2 produces a fully working, self-contained animated pelican on a bicycle, spokes, pedals, motion and all, with no broken elements. The same opossum prompt that GLM-5.1 handled brilliantly produces something nearly unrecognizable in 5.2, with no animation attempted. That regression, on a model that otherwise tops the leaderboards, is the detail that makes this release interesting rather than just impressive.
[READ ORIGINAL →]