GLM-5.2 tops open-weights AI leaderboard

Z.AI released GLM-5.2 this week, and independent testing from Artificial Analysis ranks it as the strongest open-weights model available. It scored 51 on the Artificial Analysis Intelligence Index (version 4.1), first among open models and fourth overall, behind only Claude Fable 5 at 60, Claude Opus 4.8 at 56, and OpenAI’s GPT-5.5 at 55.

The model is a large mixture-of-experts system, about 744 billion total parameters with 40 billion active per token, the same size as GLM-5.1 but with the context window expanded fivefold, from 200,000 tokens to 1 million. Z.AI publishes the weights under an MIT license on Hugging Face and ModelScope with no regional limits, and reports coding results that include 62.1 on SWE-bench Pro. The weights run on common open frameworks such as vLLM and SGLang, and Z.AI also offers the model through its own chat platform and a paid GLM Coding Plan.

Why this matters

For most of the past two years the best models were closed and metered, and the open alternatives trailed a clear step behind. A free-to-download model that lands fourth on the same index as the frontier labs narrows that gap to something a working operator can act on.

This does not mean you should rip out your current tools. It means the open option is now worth a real test, especially if you are nervous about pricing or being locked in. OpenAI’s reported losses are a reminder that paid vendors keep changing their terms, and an open model you can host or move is a genuine exit ramp. The earlier Kimi K2.7 open coding model was the same trend in smaller print, and the same care you would bring to picking any tool applies here too.