The powerful new multi-modal model boosts reasoning, code generation, and real-time inference, and will be available to both free and paid users.