Meta has announced its most ambitious open-source AI models to date with the release of the Llama 4 AI family—Scout, Maverick, and Behemoth. The surprise announcement marks Meta’s move to regain leadership in a rapidly evolving AI environment. The models provide robust capabilities in text, images, and video, and are designed to take on competitors such as OpenAI, Google, and DeepSeek. But the rollout is not controversy-free, particularly with Meta restricting access in the EU due to stringent AI regulations. With the world AI race intensifying, Meta’s Llama 4 move has the potential to change the dynamics of open AI.
Maverick and Scout Show Power
Maverick, Meta’s new leader, promises to surpass GPT-4o and Gemini 2.0 in coding, reasoning, and visual tasks. Scout is less massive but equally impressive. It has 17 billion active parameters and a 10-million-token context window, making it a breeze to work with long scholarly and legal texts—even on one Nvidia H100 GPU.
Nevertheless, Meta has had access limited in the European Union. EU companies and developers can’t distribute or use Llama 4 AI. Meta attributes stifling innovation to tough regulatory environments. Already, the move has been the subject of controversy between policymakers and tech bosses.
Behemoth Backs Meta’s Comeback
Still in the works, Behemoth is Meta’s most ambitious AI model to date. With 288 billion active parameters and close to 2 trillion total, it’s one of the largest ever constructed. Initial tests indicate that it may surpass GPT-4.5 and Claude 3.7 Sonnet on challenging STEM tests. Yet Gemini 2.5 Pro remains ahead in a number of important areas.
Meta accelerated Llama 4 development following DeepSeek’s success, which revealed Llama 2’s limitations. Internal “war rooms” were initiated to reverse-engineer their success. Now that Scout and Maverick are live on platforms such as Hugging Face, everyone is waiting for Behemoth’s last performance.
Llama 4 AI may signal Meta’s return to the top—and redefine the future of open-source artificial intelligence.