For anyone who actually has a ChatGPT subscription and relies on it, there are lots of advantages to running open weights models instead, even if you don’t have the hardware for them yourself and need to go through some provider. The privacy angle I’m sure needs no explanation, but there’s also reliability to consider. With services like ChatGPT they change how it works in the background without telling you and without any real way to opt out, so if you have a workflow or even worse an application that uses its API, what worked in the past may not work the same in the future and you have zero control over it. For overall performance the closed models are still on top, but not by that much, open models are competitive and even stuff that will run on lower powered hardware is good enough for lots of things you might need it for. Personally I don’t need anything that won’t run on my 3090.