GPT-4 is a large multimodal model (accepting text or image inputs and outputting text) that can solve difficult problems with greater accuracy than any of OpenAI's previous models, and the turbo variant is optimized for quick responses and natural language chat.