Meta: Llama 4 Scout

Meta: Llama 4 Scout 17B Instruct (16E) is a cutting-edge mixture-of-experts (MoE) language model developed by Meta. It activates 17 billion parameters out of a total of 109B, offering robust performance for various AI tasks. This model uniquely supports native multimodal input, allowing it to process both text and image data seamlessly. It also provides multilingual output across 12 supported languages, making it highly versatile for global applications. Designed specifically for assistant-style interaction and advanced visual reasoning, Llama 4 Scout utilizes 16 experts per forward pass. It boasts an impressive context length of 10 million tokens and was trained on a massive corpus of approximately 40 trillion tokens. Built for high efficiency and suitable for local or commercial deployment, it incorporates early fusion for seamless modality integration. The model is instruction-tuned for multilingual chat, captioning, and complex image understanding tasks. Released under the Llama 4 Community License, it was last trained on data up to August 2024 and launched publicly on April 5, 2025. Access Llama 4 Scout for FREE on Multi AI. It offers a context window of 327K tokens and a max output of 4K tokens. Pricing is competitive at $0.08 per 1M input tokens and $0.30 per 1M output tokens. Capabilities include vision, functions, code, and streaming, making it ideal for analysis and document processing. Note that it does not support image generation.