As the tech world expands daily, various artificial intelligence (AI) firms are launching new and advanced innovations. Recently, the US tech giant Meta has launched its latest open-weight large language model (LLM), Llama 4.
The firm claims that its newly launched model Llama 4, is an open-weight answer to its rival firms like DeepSeek, Google, OpenAI, etc.
What is the Llama 4 AI model launched by Meta?
Llama 4 is a new set of AI models that will make Meta AI smarter on apps like WhatsApp, Instagram, and Messenger. Two models, Llama 4 Scout and Llama 4 Maverick, can be downloaded from Meta’s Llama website or Hugging Face. They also showed off Llama 4 Behemoth, a special model that helps train the other Llama 4 models.
Simply put, publicly available models trained with weights are known as open-weight AI models. Further, the users can also run them locally without depending on cloud APIs. While available, the models may still have licensing limitations on modifications or commercial use.
According to reports, Meta designed Llama 4 Scout, a smaller AI model, to be natively multimodal. It uses a mixture-of-experts (MoE) design, which is like having a team of 16 specialized experts who work together, with 17 billion active parameters to make it smart.
It can process a huge amount of information at once, up to 10 million pieces of data like words or bits of images. Plus, it’s efficient enough to run on just one powerful Nvidia H100 GPU, which is a type of computer chip designed for heavy AI tasks.
Llama 4 can be your best mathematics teacher
Llama 4 Behemoth is the biggest and strongest model in the Llama 4 family, acting like a teacher to help train the smaller models. It’s also multimodal, which can handle text, images, etc.
It has 288 billion active parameters and a massive total of nearly two trillion parameters overall. Meta claims it’s a top performer in things like math, working with different languages, and image tasks, making it perfect for guiding the other Llama 4 models.
Additionally, Meta also claimed that its new AI models can beat Google’s Gemma 3, Gemini 2.0 Flash-Lite, OpenAI’s ChatGPT-4o, and Mistral 3.1 in various tests that measure how well AI models perform.
Further, the US tech giant Google’s CEO, Sundar Pichai, has also shared an X post congratulating Meta’s team on the launch of the Llama Model.
Also Read: o3 and o4-mini To Be Launched Soon, GPT-5 to Follow: Sam Altman
