Meta releases Llama 4 with Mixture of Experts architecture

Meta has released Llama 4 in April 2025, featuring a new Mixture of Experts (MoE) architecture. Two variants, Scout and Maverick, are available, with Scout serving as a balanced default and Maverick offering broader knowledge for specialized tasks. Both models leverage MoE to activate approximately 17 billion parameters per token, enabling high performance comparable to much larger models while remaining runnable on consumer hardware. AI

IMPACT Sets a new standard for locally runnable large models, potentially accelerating adoption of advanced AI capabilities on consumer hardware.

RANK_REASON New model release from a frontier lab (Meta) with a new architecture (MoE). [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on dev.to — LLM tag →

• DeepSeek-R1
• Llama 4
• Maverick
• Meta
• Mixture of Experts
• Ollama
• Qwen
• RTX 4090
• Scout

• model release
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →