Meta Unveils Llama 3.1, Largest Open-Source AI Model

Meta has released Llama 3.1, the largest-ever open-source AI model, which the company claims outperforms GPT-4o and Anthropic’s Claude 3.5 Sonnet on several benchmarks. This release follows Meta’s promise in April to create an open-source model matching the best private models from companies like OpenAI.

Llama 3.1, with 405 billion parameters, was trained using over 16,000 of Nvidia’s expensive H100 GPUs. Despite the high development costs, Meta is offering Llama 3.1 for free, requiring only approval from companies with hundreds of millions of users. In a blog post, CEO Mark Zuckerberg argues that open-source AI will soon surpass proprietary models, similar to how Linux became the dominant open-source operating system.

Meta’s collaboration with over two dozen companies, including Microsoft, Amazon, Google, Nvidia, and Databricks, aims to help developers deploy their own versions of Llama 3.1. Meta claims the model costs about half as much to run in production compared to OpenAI’s GPT-4o. The model weights are being released so companies can train it on custom data and tune it to their needs.

Llama 3.1 uses synthetic data to enhance its performance, with Meta’s VP of generative AI, Ahmad Al-Dahle, predicting its popularity as a teaching tool for smaller models deployed cost-effectively.

For the first time, Meta’s red teaming of Llama 3.1 included tests for potential cybersecurity and biochemical use cases. The model can integrate with a search engine API to retrieve information from the internet and execute tasks, such as generating Python code to plot data.

Meta’s AI assistant, integrated into Instagram, Facebook, and WhatsApp, will now feature Llama 3.1. Initially available through WhatsApp and the Meta AI website in the US, the assistant will soon support new languages, including French, German, Hindi, Italian, and Spanish. The advanced 405-billion parameter model will be accessible for a limited number of prompts each week, after which the assistant will switch to the scaled-back 70-billion model.

The “Imagine Me” feature in Meta AI scans users’ faces to insert their likeness into generated images, aiming to avoid deepfake creation by not using profile photos. Meta AI will also be integrated into the Quest headset, replacing its voice command interface and allowing users to identify and learn about real-world objects while in passthrough mode.

Meta’s predictions indicate that its AI assistant will become the most-used chatbot by the end of the year, surpassing ChatGPT’s 100 million users. Despite the hype around AI, Meta and other industry players believe this is just the beginning of the race.