Meta has unveiled its next-generation Llama 4 AI models, introducing three variants: Scout (109B parameters), Maverick (400B parameters), and the upcoming education-focused Behemoth. While Scout and Maverick are now available across Meta’s apps including WhatsApp, Messenger, and Instagram, early testing shows the technology still trails behind industry leaders ChatGPT and Gemini in key areas.
The new models employ an advanced ‘mixture of experts’ training approach inspired by DeepSeek’s innovations. Meta claims the forthcoming Behemoth version already outperforms GPT-4.5 and Gemini 2.0 Pro in STEM benchmarks, though it remains in development.
Despite these advancements, Meta AI continues to lack several features users have come to expect. The platform still doesn’t support direct image uploads for analysis, instead requiring URL-based image processing. It also falls short in AI search capabilities, deep reasoning functions, and image generation quality compared to OpenAI’s latest offerings.
While maintaining its position as a leader in open-source AI development, Meta’s consumer-facing implementation appears to prioritize broad accessibility over cutting-edge performance. The company has made all models available for download on Hugging Face and its own platforms, encouraging developer experimentation even as it works to close the feature gap with competitors.
For more breaking tech news and analysis, visit DC Brief.
