Meta Releases Llama 4 Scout and Maverick Models

Summary

Meta released Llama 4, introducing "Scout" and "Maverick" model tiers with a new Mixture-of-Experts architecture and native multimodal capabilities. The release continued Meta's strategy of open-weight releases to commoditize the model layer, while pushing the frontier of what open models could achieve.

What Happened

On April 5, 2025, Meta released Llama 4 in two initial configurations: Llama 4 Scout (a smaller, efficient model designed for broad deployment) and Llama 4 Maverick (a larger, more capable model for demanding applications). Both models used a Mixture-of-Experts architecture and featured native multimodal capabilities, processing text, images, and video.

Llama 4 Scout was designed to run efficiently on a single GPU, making it accessible for individual researchers and smaller organizations. Maverick targeted enterprise and research applications requiring higher capability, with performance competitive with leading proprietary models.

Meta emphasized the models' improved multilingual support, enhanced reasoning capabilities, and stronger performance on agentic tasks. The models were released under an updated Llama Community License that maintained commercial use permissions.

Why It Matters

Llama 4 continued Meta's evolution from a social media company to one of the most significant contributors to open-source AI. The Mixture-of-Experts architecture and multimodal capabilities brought the Llama family closer to architectural parity with proprietary models from OpenAI and Google, narrowing the capability gap between open and closed models.

The continued investment in open-weight releases reinforced Meta's strategic bet that commoditizing the AI model layer benefited the company more than restricting access. Each Llama generation raised the floor of freely available AI capability, making it harder for API-only providers to charge premium prices for similar performance.