If text-gen got new toys at the speed image-gen and video-gen do, they'd probably have triple the max context size and amazing prose in even the crappiest models. Then again the market is probably 10000 times smaller.