Meta released Llama 4, comprising open-source Scout and Maverick models alongside a proprietary Behemoth variant framed as a teacher model. All three are multimodal systems processing text, video, images, and audio. The announcement includes an implicit competitive positioning statement: Meta acknowledges Llama 4 underperforms OpenAI in humanlike voice conversation, a capability parity gap. Meta commits $65 billion in 2025 AI infrastructure capex amid investor pressure to demonstrate ROI. The announcement reflects both open-weight model advancement and the durable commercial relevance of frontier proprietary variants, while surfacing a specific capability boundary where leading closed models retain advantages.
214 other tracked developments touching the same claims.