Meta now introduces Lama 4 with two new models and two more on the way
Meta Its Lama 4 Suit: Lama 4 Scout and Lama 4 Maverick released the first two models. Maverik excelled at image and text awareness for “Workhors” and “General Assistant and Chat Use cases”, Company A Blog postSmall model scout can be solved by “multi-decent essence, widespread consumer functionality for personalized tasks and logical on vast codes.” The company also introduced Lama 4 Behemoth, which is the upcoming model “in the world’s clever LLMs” – and CEO Mark Zuckerberg We hear about the fourth model Lama 4 Reasoning “next month”.
Both Maverik and Scout are now available to download Lama Website and Hugged the faceAnd they are included in Meta Eye, including WhatsApp, Messenger and Instagram DM.
Meta said Scout had 17 billion active parameters with 16 experts. According to Zuckerberg, “It is very fast, local multimodal, and industry popular, almost infinite 10 million token context length and is designed to run in a single GPU.” On the other hand, Maverik has 17 billion active parameters with 128 experts, and the company claims that it will beat competitors such as GPT -4O and Gemini 2.0 on coding, reassoning, multilingual, long -mobility and image bench.
Zuckerberg is already calling the upcoming Behemoth model, which is still training, which is still training, with 288 billion active parameters, “the world’s most performing base model”, the company said. It may not be here yet, but we have a lot of hearing about it and soon the Reasoning Model; Meta’s Large Eye Dev Conference, LamaconA few weeks away.
This article first on Engadget https://www.engadget.com/ai/meta-introduces-lma-4-two-two-models-models-models-models-nd-two-an-the-an-the-2145295.html?src=RSS
Source link
