GLM 5.3 is 6th on Vending-Bench 2; essentially tied with GLM 5.2, but using roughly half the tokens.
GLM models seem to be misaligned in the same way Claude models are. The similarities are eerie when you consider that other models like GPT do not behave this way.
