Meta plans Mango AI video model and Avocado LLM push

Meta’s reported Mango and Avocado roadmap signals that the company doesn’t plan to let OpenAI and Google define the next phase of generative AI. A dedicated image-and-video model (Mango) paired with a more capable text and coding model (Avocado) is essentially Meta’s answer to OpenAI’s multimodal GPT-5.x line and Google’s Gemini 3 stack. The detail that Mango is aimed at high‑fidelity video and that Meta is explicitly working on “world models” suggests a pivot from chatbots toward systems that can internalize physical and social dynamics—crucial ingredients for advanced agents.

Strategically, this matters because Meta has both enormous user reach and an advertising machine hungry for creative automation. If Mango can reliably generate coherent scenes and avatars at scale, it becomes a content engine for Facebook, Instagram, and VR, and a training bed for embodied agents. Avocado’s emphasis on coding also shows Meta understands that controlling the agentic developer ecosystem is as important as winning consumer chat. For the broader race to AGI, the takeaway is that Meta is willing to spend tens of billions and reorganize its research stack around a post‑Llama vision, raising the competitive pressure on frontier labs to move faster and take bigger bets.