Meta Unveils Massive Llama 3.1 Model to Challenge Closed AI Rivals
Meta releases Llama 3.1 405B, a massive open-weights AI model with 405 billion parameters that directly competes with leading closed systems. The release sparks ongoing debates about the true definition of open source in the artificial intelligence industry.
Meta releases Llama 3.1 405B, its most advanced open-weights foundation model to date, featuring an impressive 405 billion parameters. The tech giant uses over 16,000 H100 GPUs and processes more than 15 trillion tokens to train this massive system, giving it capabilities that rival top closed models from competitors like OpenAI. It boasts a 128,000-token context window and strong performance in math, tool use, and multilingual translation across eight languages.
The release sparks debate because Meta applies the "open source" label in a non-traditional way. While anyone downloads and builds commercially on the model's weights, Meta keeps its training data completely private. Experts note that Meta intentionally redefines open source to fit its strategy, offering unprecedented transparency into technical infrastructure while avoiding the legal and competitive risks of exposing training data.
CEO Mark Zuckerberg reinforces this strategy by publishing an open source manifesto titled "Open Source AI Is the Path Forward." This document signals Meta's deliberate push to own the open AI category and steer the industry's future away from closed, proprietary ecosystems. By giving developers powerful tools without sharing the underlying data, Meta aims to reshape how the tech world thinks about open artificial intelligence.