Llama 4 Scout 17B-16E-Instruct is a 17B-parameter, instruction-tuned multimodal model developed in collaboration with AI Kosh, Meta and Sarvam. Built on a 16-expert Mixture-of-Experts architecture, it is optimized for high-performance text generation and image understanding, delivering strong efficiency and scalable multimodal reasoning capabilities.
Meta Llama 4 Scout 17B-16E-Instruct is part of the Llama 4 family of natively multimodal large language models, developed in collaboration with AI Kosh, Meta and Sarvam Built on an auto-regressive transformer architecture with a 16-expert Mixture-of-Experts design, it is optimized for strong efficiency and high-performance inference. The model is instruction-tuned for assistant-style interactions, advanced natural language generation, and multimodal reasoning tasks including image understanding, captioning, and visual question answering. Released on April 5, 2025, Llama 4 Scout supports a broad set of languages such as English, Hindi, French, German, Spanish, Arabic, Indonesian, Italian, Portuguese, Tagalog, Thai, and Vietnamese, and is distributed under the Llama 4 Community License for both commercial and research use.
Llama 4 community license
Sarvam AI
Conversational AI Model
Transformers
Open
Sector Agnostic
15/02/26 16:04:16
162.05 GB
Llama 4 community license
© 2026 - Copyright AIKosh. All rights reserved. This portal is developed by National e-Governance Division for AIKosh mission.