Meta Llama 4 Scout and Maverick — Open-Weight MoE, 10M-Token Context, and the Benchmark Controversy

Meta Llama 4 Scout and Maverick launched April 5, 2025 with Mixture-of-Experts architecture, native multimodal early fusion, 10M-token context (Scout), and aggressive pricing from third-party providers. We review the architecture, benchmarks, the LMArena controversy, licensing, and what open weights actually mean in 2025.