Yeah, all of OpenAI's stuff gets better much quicker than I'm used to. So does the general performance ceiling of all open source models, even if individual models don't improve as much.