Llama 4 Scout 17B Instruct (US)

Llama 4 Scout 17B Instruct is the instruction-tuned member of Meta's Llama 4 family, positioned as a multimodal conversational model with open weights that can be self-deployed or run through managed providers. It accepts both text and image prompts while returning text responses, which makes it suitable for assistants that need to reason over documents, screenshots, or mixed visual-text inputs alongside standard chat. The model is explicitly built to be developer-friendly, exposing tool calling, attachment handling, and temperature control for shaping responses, and its open-weights status means organizations can inspect, fine-tune, or run the weights on their own infrastructure rather than depending solely on hosted endpoints.

Where Llama 4 Scout stands out is in its exceptionally large context budget, paired with a modest 16K output ceiling that fits typical response-length workflows. That combination makes it a natural fit for long-document retrieval-augmented generation, multi-file code or log analysis, and agentic loops that need to keep large tool histories in view, while the image input capability extends those use cases to diagrams, UI screenshots, and visual Q&A. Because the weights are open, teams can also distill or specialize the model for narrower domains, and the published per-token pricing on Bedrock makes high-volume, context-heavy workloads economically tractable for production experimentation.