META launched the first two models at Multimodal Llama 4 Suite: Llama 4 Scout and Llama 4 Maverick. Maverick is said to be “The WorkHorses” of these two and excellent in understanding images and texts of “general auxiliary and chat use cases”. BlogSmall model scouts can deal with things like “inferences for vast codebase by parsing a wide range of user activities for personalized work and personalized work.” The company also introduced the upcoming model LLAMA 4 Beemoth. Mark Zuckerberg We have heard of the fourth model, LLAMA 4, “next month.”
Maverick and Scout can be downloaded from both. llama Website and HugAnd added to Meta AI, including whatsapp, messenger and instagram DMS.
Scouts have 16 experts and 17 billion active parameters, Meta says. According to Zuckerberg, “very fast and basically multi -modal, almost infinite 10 million token context lengths, and are designed to run on a single GPU.” MAVERICK, on the other hand, has 17 billion active parameters with 128 experts. The company said that it has won competitors such as GPT-4O and Gemini 2.0 in coding, reasoning, multi-language, long-term text and image benchmarks, and has accumulated against the Deepseek V3.1 for reasoning and coding.
According to the company, Zuckerberg is the “world’s highest performance and basic model”, which is still training the coming Beemoth model and has a 20 billion active parameter. It may not be here yet, but we will be more about it. META’s Big Ai Developer Conference, Llamacon, is just a few weeks away.
This article was originally in Engadget https://www.engadget.com/ai/meta- INTRODUCES-LAMA-4-WITH-WITH-WITH-Models-nod-NOD-MORE-WAY-WAY-WAY-214524295.HTML?src=rss
