Llama 4 Scout 17B Instruct (16E) is a Mixture-of-Experts (MoE) model from Meta, activating 17B parameters out of 109B total. It supports native multimodal input (text + images) and generates multilingual text and code across 12 languages. With 16 experts per forward pass, Scout is optimized for assistant-style interaction, visual reasoning, and large-scale context handling—supporting up to 10M tokens and trained on a ~40T-token corpus.
Engineered for efficiency and flexible deployment, Scout uses early fusion for smooth multimodal integration and is instruction-tuned for tasks like multilingual chat, captioning, and image understanding. Released under the Llama 4 Community License, it was trained on data up to August 2024 and made publicly available on April 5, 2025.
| Creator | Meta |
| Release Date | April, 2025 |
| License | Llama 4 Comunity Lisense Agreement |
| Context Window | 128,000 |
| Image Input Support | Yes |
| Open Source (Weights) | Yes |
| Parameters | 109B, 17B active at inference time |
| Model Weights | Click here |
