The models themselves support up to 1M. You are just charged more for all context over 400k. The 272k limit is just a client limitation to ensure you can never go over 400k since the maximum output size of the model is 128k.
replyNot true. This is the case for the API, not for subscriptions: they are hard capped at 400k.
replyPerhaps not this for this parameter, but for others? Client side.
reply