Meta AI for eCommerce. Products: Llama model weights: Meta's open-weight AI models available for download and self-hosting. You manage deployment, infrastructure, privacy, and costs. Can also run via cloud partners (AWS, Azure, GCP) without managing hardware directly.. Best for: eCommerce teams that need full data privacy, want to avoid per-token API costs at scale, or need to fine-tune a model on proprietary product data for specialised tasks.. Access: Llama 4 Scout; Llama 4 Maverick (plus pretrained variants).. Meta AI: Meta's consumer AI assistant embedded in WhatsApp, Instagram, Messenger, and the web. Powered by Llama models, it offers conversational AI within Meta's social platforms.. Best for: Quick inspiration and ad-hoc queries within Meta's ecosystem. Not typically used for business automation, but useful for testing ideas before building with Llama directly.. Access: Powered by Llama models (version depends on region/app).. Models: Llama 4 Scout (Instruct) (meta-llama/Llama-4-Scout-17B-16E-Instruct): High-volume private workloads; long-context analysis; self-hosted support. Type: Open-weight multimodal. Very long context. Needs engineering to deploy well (RAG, evals, safety). Llama 4 Maverick (Instruct) (meta-llama/Llama-4-Maverick-17B-128E-Instruct): Stronger open-weight option; internal copilots; data-private agents. Type: Open-weight multimodal. Shorter context than Scout but often stronger general performance. Llama 4 Behemoth (preview) ((not generally available)): Not for direct deployment (as of public info).. Type: Teacher model (preview). Mentioned as a preview/teacher model; treat as roadmap, not a tool you can buy today.