Meta AI Launches Llama 4: Insights, Delays, and Community Reactions

Meta has launched the first models in its Llama 4 family, Scout and Maverick, which showcase a native multimodal architecture and a mixture-of-experts framework, aiming to broaden applications from image understanding to long-context reasoning. Scout features 17 billion active parameters across 16 experts, optimized for a single NVIDIA H100 GPU with a 10 million token context window, while Maverick also has 17 billion parameters but with 128 experts, excelling in reasoning and coding. Both models were distilled from the larger Llama 4 Behemoth, noted for its 288 billion active parameters and superior performance on STEM benchmarks compared to GPT-4.5. Despite promising benchmarks, user skepticism about real-world performance has surfaced. Concurrently, Meta announced the departure of Joelle Pineau, its VP of AI Research, as the company prepares for the LlamaCon AI conference. In a parallel development, Google introduced Gemini 2.5, its most advanced generative AI model, claiming superior performance over competitors with its multimodal capabilities and extensive context window, positioning it as a powerful tool for enterprise applications on Google Cloud's Vertex AI platform.