per task versus score which is you know usual cost versus capabilities on terminal bench science. I I know the the organizer of that uh and deep suite and what you like to see is models that are getting less expensive for this for constant capabilities and that's exactly what we're seeing here. So one can reasonably suspect that GPT 6.1 SOL is some sort of distillation of GPT6 Astra or some common root model. Maybe it's
So one can reasonably suspect that GPT 6.1 SOL is some sort of distillation of GPT6 Astra or some common root model.