Llama 4 Maverick is Meta's high-performance multimodal language model that achieves GPT-4o and Gemini 2.0 Flash-level performance with a mixture-of-experts architecture using 17 billion active parameters from 400 billion total. Maverick excels at complex reasoning, coding, and visual understanding tasks, matching or exceeding closed-source frontier models while being fully open-source. It supports interleaved image and text inputs with a 1 million token context window, enabling sophisticated multimodal analysis at a level previously only available through expensive API calls.