GLM-5.2 tops Artificial Analysis Intelligence Index
Z.ai has released GLM-5.2, a 744B parameter mixture-of-experts model scoring 51 on the Artificial Analysis Intelligence Index v4.1 to lead all open-weights models. The MIT-licensed model expands the context window to 1M tokens and matches GPT-5.5 on agentic workflows, scoring 1524 on the GDPval-AA v2 benchmark.
GLM-5.2 establishes a new frontier for open-weights agentic performance, matching proprietary giants on complex workflows, though it achieves this by consuming a high volume of output tokens for reasoning.
* Significant improvements over GLM-5.1 across evaluations, led by CritPt (+16 points), HLE (+12 points), and TerminalBench v2.1 (+16 points).
* Excellent cost-efficiency, sitting on the Intelligence vs. Cost per Task Pareto frontier at ~$0.46 per task.
* High token usage, consuming 43k output tokens per task (with 37k reasoning tokens), which makes it less token-efficient than peers like DeepSeek V4 Pro.
* MIT license and broad API provider availability (e.g., DeepInfra, Nebius, Fireworks) enable rapid adoption and deployment flexibility.
DISCOVERED
4d ago
2026-06-17
PUBLISHED
4d ago
2026-06-17
RELEVANCE
AUTHOR
himata4113