From what I've read online it's not necessarily a unquantized version, it seems to go through longer reasoning traces and runs multiple reasoning traces at once. Probably overkill for most tasks.