They should also at least run Opus through the same Pydantic harness they used for GLM. As is, it's apples vs pears.

Where's the cost per vulnerability for all the other models than GLM?

Also, without code this isn't very trustworthy. Could all be made up as well.