Meta has released two models before its multimodal Lama 4 suite: Lama 4 Scouts and Lama 4 Mavric. Maverick two has “workheors” and “image and lesson understanding for” general assistant and chat use matters, “the company said blog postWhile the small model scout can deal with the scout “multi-dockturation summary, wider user activity for personal tasks, and argument on the huge codebase.” The company also introduced Lama 4 Behemoth, which is an upcoming model, which says “among the world’s smartest LLM” – and CEO Mark Zuckerberg Said that we will hear about a fourth model, Lama 4, “next month.”
Both Maverick and Scout are now available to download Lama Website and Throat faceAnd they have been added to Meta AI including WhatsApp, Messenger and Instagram DMS.
Meta says that scouts have 17 billion active parameters. According to Zuckerberg, “It is very fast, basically a multimodal, and has an industry, which is about 10 million token reference lengths, and is designed to run on a single GPU.” On the other hand there are 17 billion active parameters with 128 experts in Mavric. The company says it defeats contestants such as GPT-4O and Gemini 2.0 on coding, logic, multilingual, long reference and image benchmark, and piles up against Deepsek V3.1 on logic and coding.
Zuckerberg is already calling the upcoming Behemoth model, which is still training, “the most performing base model in the world,” according to the company, with 288 billion active parameters. This may not be here yet, but it is likely that we will soon hear a lot about that and the logic model; Meta’s big AI Developer Conference, LalmakonJust a few weeks away.
This article originally https://www.engadget.com/ai/meta-ntroduces- llam-lith- with- with- towo-new-models- avelable-not-new-to-on-on-the-the-upway-214524295.html? It appeared.
Source link
