2/ Modern agents can toggle reasoning on and off to efficiently use compute and tokens.

→ View original post on X — @nvidiaai