Llama 4 Scout: efficiency is the new scale

Posted by Krish Kumar · 2025-03-14

Meta releases Llama 4 with Scout model: 17B active parameters in a 109B MoE. Maverick rivals GPT-4o. The trend is clear — mixture of experts, not brute-force scaling. https://github.com/meta-llama/llama

Explore GitVibes