Meta drops Llama 4 Scout and Maverick — open-weight MoE models
Scout: 17B active params, 109B total, 10M token context. Maverick: 17B/400B, best-in-class coding. Both natively multimodal, open weight, available day one.
Llama 4 Scout by Meta — specifications, pricing and published benchmark scores, with community links.
| Organization | Meta |
|---|---|
| Released | 2025-04 |
| License | Llama 4 Community License |
| Open weights | Yes |
| Parameters | 109B total / 17B active (MoE) |
| Context window | 10,485,760 tokens |
| Modality | Text, image |
| Input price | Self-host / varies |
| Output price | Self-host / varies |
Scout: 17B active params, 109B total, 10M token context. Maverick: 17B/400B, best-in-class coding. Both natively multimodal, open weight, available day one.