The latest family of the Meta Generator AI model unveiled Saturday. You can give Lama 4 Models are now a test drive through the Meta AI website, and Lama 4 will soon give the company’s Instagram, WhatsApp and Messenger services many meta AI features.
The competition between meta and other AI companies is becoming increasingly intense. Companies are working to create and publish AI models capable of more complex work and advanced logic so that there is no need to run a lot of computing energy and cash. It is a complex sweet spot to hit, and the meta hopes that its new models will keep it ahead of competitors like Chatzipt and Jemini.
There are now two models in the Lama 4 family: Scouts and Mavarick. These are open-wail models and multimodals, which means they can create text, images and codes. Open models like Meta’s meaning mean that developers can get some insight about how the models are built. The Lama 4 models are open-way models, which means you can see how the model is connected and how to learn the specific features more weight is given. Open announced earlier this month that it was developing an Open Open Model for the first time.
The scout is the smallest model of the family, it is designed to run a single Nvidia H 100 GPU. The scout has a 10 million token context window and it is a 17 billion parameter model that features 16 experts (subnetworks in models, allowing it to run more efficiently). Scout Lama 3 has more than twice the firepower, which has 8 billion parameters. Generally, the more parameters of a model have, the better to provide better results.
Mavarick is a midsized model, scout to scout, with 128 experts featured 17 billion parameters. Benchmark tests have shown that the Mavarick DIPSEC V3 has shown that the textGPT -4O and the text generation have defeated the generation. In the case of logic and coding, the dipsek still has the edge. CNET did not independently verify the Mater Benchmarking tests.
More information about the rest of the Lama 4 family with a base model called Behemoth and a Lama 4 argument model is expected to come to the end of this month, CEO Mark Zuckerberg posted according to a videoThe We will probably learn more about these models at the company’s first annual AI developer conference in Llamacon, starting April 25.
For more information, we see what we know about our review of a potential Standelone app and best AI chattobs for Meta AI.