Meta released its Llama 4 models, headlined by Scout and Maverick, the first open-weight natively multimodal models in its lineup.
The company also previewed Llama 4 Behemoth, billed as one of the smartest large language models in the world.
Meta unveiled the models at LlamaCon, its first event dedicated to the Llama developer community.
Meta Llama 4 Scout and Maverick Models
Llama 4 Scout is a 17 billion active parameter model built with 16 experts.
Meta called Scout the best multimodal model in the world in its class.
It offers an industry-leading context window of 10 million tokens.
Llama 4 Maverick uses 17 billion active parameters with 128 experts.
As Meta stated, Maverick beats GPT-4o and Gemini 2.0 Flash on many benchmarks.
Llama 4 Behemoth and the Architecture
Meta previewed Llama 4 Behemoth as its most powerful model yet.
Behemoth is designed to act as a teacher that trains Meta’s smaller models.
The Llama 4 models are the first Meta built using a mixture-of-experts architecture.
That design activates only part of the model per task, improving efficiency.
LlamaCon and New Safety Tools
LlamaCon brought together developers from around the world who build with Llama.
Meta released new protection tools for the open source community at the event.
These included Llama Guard 4, LlamaFirewall, and Llama Prompt Guard 2.
The tools aim to help developers deploy Llama models more safely.
Meta Plans an AI Prediction Market App
Meta plans to release an AI-powered prediction market app, documents show.
The app, codenamed Antwerp and FBForecast, would use Llama to generate questions.
It would pull trending topics automatically and make personalised market recommendations.
The plan signals Meta weaving Llama deeper into consumer products.
The open-weight push contrasts with rivals, as seen in our OpenAI hardware coverage.
As Meta announced at LlamaCon, Llama remains central to its open AI strategy.
The mixture-of-experts design marks a shift for Meta’s flagship models.
Scout’s 10 million token context window is among the largest available.
Meta positioned the open-weight models against closed rivals from OpenAI and Google.
Developers can build on Llama without the licensing limits of some competitors.
The safety tools reflect growing scrutiny of open AI models.
Behemoth is still in preview rather than a public release.
Meta has invested heavily in AI infrastructure to train these models.
The prediction market app would resolve outcomes using AI.
Llama 4 strengthens Meta’s pitch to the open source community.
Analysts watched closely to see how Llama 4 compares in real use.
The models support text and images in a single native system.
Meta continues to release frequent updates across its AI portfolio.
The launch deepens competition at the frontier of artificial intelligence.
Meta reaffirmed its commitment to open-weight AI models.
Developers can build freely on the Llama 4 family.
The release intensifies competition at the AI frontier.
Meta continues to invest heavily in AI infrastructure.
Related Articles
Anthropic Accuses Alibaba of Illicitly Extracting Claude AI Data
Google Gemini 2.5 Pro Update Brings Deep Think and New AI Tools
OpenAI and Broadcom Unveil Jalapeno, Their First Custom AI Chip
Trust Post Desk
A journalist and editor at TrustPost.org covering world and national news, technology updates and human-interest stories. They check every fact, interview sources in person or online, and aim to deliver clear, accurate reporting. Their work ranges from breaking news to in-depth features and daily newsletters. Outside the newsroom, they follow emerging trends and engage with readers on social media.
