I think Llama 4 scout/maverick (forgot which one) has 10 million tokens, which is a huge upgrade over Llama 3.3's 128k tokens, so it went from 4k > 128k > 10m. as I said, the progress of Ai is terrifying, the only thing they need to do right now is to work on power efficiency because these models require at least five nuclear power plants to function properly. I heard somewhere that Llama 4 scout could be the new text model, but of course it won't have the 10m tokens, because that demands massive computational power that only mega corporations could pull off, so if the rumors are true and it's actually the chosen model, it will probably tuned down to 128k/200k, depending on the dev's hardware and how he's going to deal with it, if the rumors surrounding Llama 4 scout are a bunch of bull, then mistral small 3.1 or Llama 3.3 will be chosen instead.