Anyway I think if you have a single stream of a cheap model, like GPT 6 Luna, I don't think you can currently exhaust it in a week on a $200 plan. I mean it only puts out so many tokens per second.
Is it just the benchmarks? Because otherwise it suggests it's twice as chatty as Opus for a comparable output... Which kind of defeats the purpose
In xhigh effort it is a lot cheaper and possibly lot less impressive?
I'm very curious how do they know what requests could assist competing AI models.