Meta Unveils the Original "LLaMA": A Research-Oriented Foundation Model Enters the Generative AI Race
On February 24, 2023, Meta announced “LLaMA (Large Language Model Meta AI),” a foundation language model. Coming just three months after ChatGPT’s debut, this marked Meta’s first step in releasing its own large language model to the research community — and the origin point for the entire “Llama” series that followed.
Details
- Four sizes released: Four models were released simultaneously — 7B, 13B, 33B, and 65B parameters. Meta stated that the largest 65B model could compete with far larger models such as DeepMind’s Chinchilla-70B and Google’s PaLM-540B
- Enabling research with fewer resources: Designed to achieve strong performance even at smaller scale, aiming to make large language model research accessible to researchers without access to massive infrastructure
- License: Released under a non-commercial, research-only license that prohibited commercial use, granting access through an application process to academic researchers, government, civil society, and academia-affiliated organizations, as well as industry research labs
- How it was released: The model weights themselves were not made publicly available but distributed through an application process. Within weeks, the weights leaked online, ending up widely distributed regardless
What happened next
This closed, research-only restriction did not last long — in July 2023, “Llama 2” arrived, available for commercial use by anyone. This original LLaMA became the starting point of Meta’s own pivot toward establishing itself as a champion of open-weight models.