[vmg12]

I tried 4.5 which i thought was the best model, seems like the reasoning models do get it.