AI news

1. Meta Releases Llama 4: The First Open--Weights Native Multimodal Model

Date: August 19, 2026

Meta has officially released Llama 4, its next--generation open--weights model. The model introduces "Native Multimodal Architecture," allowing it to process and generate text, image, audio, and video within a single unified transformer block.

Feature Specification
Parameters 2 Trillion (Dense)
Context Window 1,000,000 Tokens
Modality Native Text, Image, Audio, Video
License Llama Community License

Key Facts & Trends:

Source: https://ai.meta.com/blog/llama--4--release/