Meta has launched its fourth-generation Llama artificial intelligence models, unveiling three new versions — Scout, Maverick, and Behemoth — as part of a global push to accelerate its presence in the open-source AI ecosystem.

The models, introduced on Saturday, are built using a mixture-of-experts (MoE) architecture aimed at boosting efficiency by routing queries through specialised model segments.

Meta claims the suite delivers high performance in tasks involving reasoning, coding, and multimodal content such as images and videos.

Maverick, positioned as a general-purpose AI assistant, reportedly surpasses OpenAI’s GPT-4o and Google’s Gemini 2.0 in several benchmark tests.

However, it lags behind more advanced models such as GPT-4.5 and Gemini 2.5 Pro, according to TechCrunch.

Scout and Maverick are now freely accessible via Meta’s website and through distribution platforms like Hugging Face.

However, their availability excludes companies and developers based in the European Union. Meta has cited concerns over the EU’s strict AI regulatory regime, which it argues hampers innovation.

The company has been openly critical of the EU’s Artificial Intelligence Act, calling its compliance burdensome.

The latest rollout signals Meta’s intent to maintain global leadership in AI while navigating increasingly complex regulatory landscapes.

Meta’s move adds further momentum to the race between major tech firms to dominate the next wave of generative AI, particularly in open-source deployments.

The launch also underlines ongoing tensions between AI innovation and privacy regulation as companies seek to scale technologies without falling afoul of legal frameworks.