Meta has announced the release of the first two products in its highly anticipated multimodal Llama 4 suite: Llama 4 Scout and Llama 4 Maverick.
The launch marks a major step in Meta’s effort to redefine the capabilities of large language models (LLMs) across diverse applications.
At the core of the new offerings is Llama 4 Maverick, which Meta describes as the foundational model in the suite.
With extensive domain knowledge and robust reasoning capabilities, Maverick is designed to support a wide range of use cases, including general assistance and conversational tasks.
In a blog post detailing the launch, Meta revealed that a third product, Llama 4 Behemoth, is on the horizon.
The company claims Behemoth will be “one of the most intelligent LLMs in the world,” signaling a bold escalation in the AI race.
CEO Mark Zuckerberg announced the company will present Llama 4 Reasoning to the public in the following month.
Users can access Maverick and Scout through the Llama website and Hugging Face as well as Meta AI which supports these models for WhatsApp Messenger and Instagram direct messages.
Scout consists of 17 billion active parameters with sixteen experts according to Metas statements. The latest version from Zuckerberg delivers an “industry-leading 10 million token context length with native multimodal capability” at “near-infinite speed for a single GPU platform,” according to him.
The experts in Maverick number 128, while its active parameter count stands at 17 billion. Meta reports thatawahich is comparable to GPT-4o and Gemini 2.0 on coding, reasoning, multilingual, long-context and image benchmarks, stands competitively against DeepSeek v3.1 on reasoning and coding tasks.
ALSO READ
Zuckerberg declared Behemoth as the “highest performing base model in the world” due to its 288 billion active parameters while the company continues its training.
The Reasoning model, together with the upcoming Behemoth model will be major topics at LlamaCon, which is occurring in the upcoming weeks according to Meta.
