Meta Launches Lalama 4 to Rival Chatgpt Gemini Scout Maveralic Released Behemoth Under Development

Along with Meta, Mark Zuckerberg himself shared information about his latest Language Models (LLMS) with people. It consists of two major versions, the first LLAma 4 Scout Is. It is a mild model that can be run on a single Nvidia H100 GPU. It has a contact window of 10 million tokens and has a target to give high-quality output in low resources. In terms of performance, it is better than light models like Mistral 3.1 and Google Gemma 3.
At the same time, the second LLAMA is 4 Maverickk, a high-performance model, which claims to give equal or better results of advanced models like GPT-4O and Deepsek-V3. Especially in coding and logical task, its performance is said to be very impressive.
Meta is also working on a third supermodel named LLAma 4 Behemoth. It will have 2 trillion parameters and can be the largest META model ever with 288B active parameters. It has been claimed that it can also beat models like GPT-4.5 and Claude 3 Sonnet.
This time META has used Mixture of Experts (Moe) architecture in these models. The advantage of this is that not all parameters are active together, which gives fast speed and better efficiency.
META has also confirmed that its AI models are now being integrated through Meta AI in platforms like WhatsApp, Messenger, Instagram and Web. This will give users more smart and interactive chatbott experience than before.
However, this model of META cannot be called a completely open-source, as its licensing has some limitations regarding commercial use. Institutions like Open Source Initiative have also raised questions on this.
META has planned to share all these announcements at the Llamacon conference to be held on April 29.
(Except for the headline, this story has not been edited by ndtv staff and is published from a press release)
(Tagstotranslate) meta