This site is fictional demo content. It is not real news or affiliated with any real organization. Do not treat it as fact or professional advice.

NextPaper Logo
nextpaper.ccCYBER CHRONICLE
RSS
EN中文

Full article

FULL TEXT

View this issue
BriefAI

Meta Open-Sources Llama-4 with 2 Trillion Parameters

Meta open-sources the Llama-4 series with flagship version reaching 2 trillion parameters—the largest open-source model to date. Meta also releases a new open-source license prohibiting military and targeted advertising use.

Content

Meta announced on April 8th the open-sourcing of the Llama-4 series, the latest milestone in its open LLM initiative.

The Llama-4 series includes three versions: Llama-4-Scout (109 billion parameters), Llama-4-Master (2 trillion parameters), and Llama-4-Ultra (2 trillion parameters). Llama-4-Ultra is the largest open-source model to date, with parameter scale 10x that of Llama-3's largest version.

In terms of performance, Llama-4-Master approaches GPT-6 levels across multiple benchmarks, while Llama-4-Ultra surpasses GPT-6 overall, particularly in complex reasoning and multi-step tasks. Meta simultaneously released a detailed training report disclosing the model used over 15 trillion high-quality training tokens.

Llama-4 adopts a new "Magnolia" open-source license—Meta's first customized license for AI models. The license explicitly prohibits model use for military purposes, autonomous weapons control systems, and targeted advertising, sparking industry discussion. Legal experts noted the license's enforceability within the open-source community remains to be seen.

Llama-4 model weights are now available for download, with enterprise customers able to apply for commercial licensing.

Boundary

This is fictional content for entertainment only.