GLM-5.2 (max) tops AA Intelligence Index but costs 3x more than comparable open-weight models

GLM 5.2 Performance Benchmarks

Z AI released GLM-5.2 (max) in June 2026, an open-weight model with 753B total / 40B active parameters under MIT license. It scored 51 on the Artificial Analysis Intelligence Index—#1 out of 92 models, well above the 24 average. Speed is strong at 112 output tokens per second. The catch: input costs $1.40/1M tokens and output $4.40/1M tokens, roughly 3x the averages for comparable models ($0.42 and $1.25). Evaluating it on the Intelligence Index alone cost $867.88. It's also on the verbose side, generating 140M tokens vs. a 110M average. Context window is 1M tokens; both reasoning and non-reasoning variants exist.

Why it matters: GLM-5.2 tops Artificial Analysis's Intelligence Index with concrete speed and pricing data, MIT-licensed and open-weight. Score isn't higher because this is one benchmark platform's ranking, not the model release itself, and the post doesn't break down which tasks make up the ...