Meta Unveils "Llama 3.1" — the "First Frontier-Level Open Model", 405B Parameters

On July 23, 2024, Meta announced its multilingual model family “Llama 3.1.” The centerpiece was the 405-billion-parameter “Llama 3.1 405B,” which Meta positioned as the “first frontier-level open source AI model,” claiming performance rivaling top proprietary models in general knowledge, math, tool use, and multilingual translation.

Details

• Three sizes: Pretrained and instruction-tuned versions at 8B, 70B, and 405B parameters. Context length expanded to 128K tokens, and supported languages expanded to eight
• 405B training scale: Trained on over 15 trillion tokens using more than 16,000 H100 GPUs — the largest-scale training run in the Llama series to date
• Performance claims: Meta stated this was the first case of openly offering performance rivaling state-of-the-art closed models in general knowledge, steerability, math, tool use, and multilingual translation
• Availability: Downloadable from Hugging Face from launch day, and immediately available for development on a wide range of partner platforms

What happened next

The arrival of the 405B model became a symbolic moment demonstrating that open-weight models could rival closed ones. This direction carried forward into “Llama 4” in April 2025, which adopted a multimodal, mixture-of-experts architecture.