Meta launches Llama 4: Meta has launched the first batch of its most open-source AI models, Llama 4, to further strengthen its position in the field of artificial intelligence.
Meta launches Llama 4, In a videotape posted on Instagram, Meta CEO Mark Zuckerberg outlined the company’s bold AI strategy, saying that their thing is to make the world’s most advanced AI, make it open source and make it accessible encyclopedically.
According to NDTV, two Llama 4 models have been originally released, Llama 4 Scout and Llama 4 Maverick, which are available on Llama’s website and Hugging Face. These models are now also forming the base of Meta’s virtual adjunct” Meta AI”, which has been added to WhatsApp, Instagram, Messenger and the web.
Meta has also introduced the Llama 4 Behemoth model, which is being described as the most important and intelligent AI model ever. It will be used to train and guide unborn models. The highlight of Llama 4 is its new Admixture-of- Experts(MOE) frame, in which different modules concentrate on specific motifs similar as drugs, poetry, biology, and programming. Only applicable modules are active during each task, which reduces training and perpetration costs.
The Scout model is erected with 17 billion parameters and 16 experts and can be run on just one GPU. Counterculturist is also erected with 17 billion parameters but includes 128 experts, and is presented as an each-round model that can handle converse, arguments, and other digital tasks. According to Meta, Maverick outperforms several well-known models similar as GPT- 4o and Gemini 2.0 Flash, especially in rendering, arguments, and long queries.
The mammoth model, which is still in development, will have 288 billion active parameters and nearly two trillion total parameters, and is being touted as outperforming other major models similar as GPT- 4.5 and Gemini 2.0 Pro in wisdom and technology-related tests. Zuckerberg also mentioned another model, Lama 4 logic, which will concentrate on working complex logical problems. Further details will be released in the coming weeks.
Meta also revealed that downloads of Lama models have exceeded one billion, over from 650 million in December 2024. In January, Zuckerberg blazoned that Meta’s investment in AI structure in 2025 would be between$ 60 billion and$ 65 billion, which would be spent on waiters, data centers, and other coffers.