Meta has introduced two new AI models, Llama 4 Scout and Llama 4 Maverick, which are the first open-weight multimodal models featuring unprecedented context length support and built using a mixture-of-experts (MoE) architecture. However, the unmodified version of Llama 4 Maverick has recently seen its ranking decline on the LM Arena benchmark, falling below older competitors. This drop in ranking follows previous controversies surrounding the experimental Maverick model, which had utilized an unreleased version of Llama 4 to achieve high scores, prompting policy changes within the company. Meta asserts that its new models are superior to those developed by OpenAI and Google across a broad range of benchmarks.