A 13 billion parameter Llama model outperformed GPT-3, which had 175 billion parameters, on most benchmarks.

Llama 3 was released in April 2024, with an 8 billion parameter model suitable for consumer-grade devices and a 70 billion parameter model that closed the performance gap with state-of-the-art proprietary models.

Yann LeCun confirmed in a Financial Times interview that Llama 4 benchmark results were 'fudged a little bit' and that the team used different models for different benchmarks to give better results.

Meta's Llama model was downloaded over a billion times within two years, making it the most downloaded AI model in history.