Llama 4 with 3 new AI models

In Today’s Issue:

• Llama 4 releases 3 new AI models

• OpenAI banks on viral trend Studio Ghibli

Read time: 2 minutes

Meta announced the release of 3 new AI models, Llama 4 Scout, Llama 4 Behemoth, and Llama 4 Maverick. Meta’s Llama 4 models represent the company’s first step into multimodal AI, integrating text, image, and video processing into one framework. The models leverage Mixture of Experts (MoE) architecture, improving computational efficiency.

Maverick outperforms many competitors in reasoning and coding tasks, while Scout excels in large document handling. Behemoth, still under development, is aimed at advanced STEM applications with unparalleled parameter sizes. The models are expected to transform industries, enabling smarter interactions across platforms like WhatsApp and Instagram.

Technical details (for nerds):

• Scout
  Active Params: 17B
  Total Params: 109B
  Experts: 16
  Context: 10M tokens
  Hardware: 1× H100 GPU

• Maverick
  Active Params: 17B
  Total Params: 400B
  Experts: 128
  Use Case: General assistant
  Hardware: DGX-class

• Behemoth (training)
  Active Params: 288B
  Total Params: ~2T
  Experts: 16
  Focus: STEM, high-end reasoning