James Martin/CNET

Meta unveiled its latest family of generative AI model on Saturday. You can give the Llama 4 models a test drive now through Meta AI’s website, and Llama 4 will soon power the many Meta AI features on the company’s Instagram, WhatsApp and Messenger services.

The competition between Meta and other AI companies is becoming increasingly intense. Companies are working to build and release AI models capable of more complex tasks and advanced reasoning without requiring vast amounts of computing power and cash to run. It’s a tricky sweet spot to hit, and Meta hopes its newest models will put it ahead of competitors like ChatGPT and Gemini

The Llama 4 family currently includes two models: Scout and Maverick. They are open-weight models with multimodal capabilities, which means they can generate text, graphics, and code. Open models, such as Meta’s, allow developers to have insight into how the models are constructed. The Llama 4 models are open-weighted, which means you can see how the model connects and which attributes gain weight as it learns. OpenAI stated earlier this month that it is working on an open-weights model for the first time.

Scout is the smallest variant in the family, running on a single Nvidia H100 GPU. Scout has a 10 million token context window and a 17 billion parameter model with 16 experts (subnetworks that help the model run tasks more effectively). Scout has more than twice the firepower of Llama 3, which contains 8 billion parameters. In general, the more parameters a model contains, the faster it can produce better results.

Maverick is a midsized model and Scout’s older brother, with 17 billion parameters and 128 experts. According to Meta, benchmark tests showed that Maverick outperformed ChatGPT-4o and DeepSeek V3 for text production. DeepSeek continues to outperform in terms of thinking and coding. CNET has not independently confirmed Meta’s benchmarking results. According to a video produced by CEO Mark Zuckerberg, more information on the remainder of the Llama 4 family, including a base model called Behemoth and a Llama 4 reasoning model, will be released later this month. We’ll likely learn more about these models at LlamaCon, the company’s first annual AI developers conference, which starts on April 29.

For more, check out what we know about a potential standalone app for Meta AI and our review of the best AI chatbots.

.