GLM-5.2 is the new leading open weights model on Artificial Analysis

476 points · 258 comments on HN · read original →

Points and comments are a snapshot, not live.

GLM-5.2 leads open models on Artificial Analysis Intelligence Index, scoring 51.

Z ai's GLM-5.2 (744B total/40B active parameters) scores 51 on the Intelligence Index v4.1, leading open models ahead of MiniMax-M3 (44) and DeepSeek V4 Pro (44). It uses MIT license, 1M context window, and costs $1.4/$4.4/$0.26 per 1M input/output/cache hit tokens on first-party API. Gains include scientific reasoning improvements and top score on GDPval-AA v2 agentic metric (1524), matching GPT-5.5. However, it uses 43k output tokens per task, more than peers.

What commenters are saying

Commenters note GLM-5.2 is a strong model but suffers from high token usage and overthinking, especially on max reasoning setting. Some find it comparable to Opus 4.8 at far lower cost, while others warn third-party providers may misconfigure models. A user reports running a simple coding task took 15 minutes and 45k tokens; others suggest using GLM 5.2 High cuts tokens 2-2.5x with minimal quality loss. Concerns about server load and unofficial API quantization are also raised.