GLM-5.2 is probably the most powerful text-only open weights LLM (simonwillison.net)

I wonder if multiple attempts at the opossum would produce better results.

If we didn’t have the previous example I would interpret this as pretty solid evidence that labs were training on the Pelican “benchmark”.

I just can’t imagine a model dropping so significantly from one version to the next on such a silly task.

Related:

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

https://news.ycombinator.com/item?id=48567759

It’s interesting how little press Minimax M3 gets, given it outperforms Deepseek V4 Pro, previously the SOTA for open models. Meanwhile GLM has been in the news daily.