LLaMA (Large Language Model Meta AI) is a family of large language models developed by Meta AI, first released in February 2023. The models range in size from 1 billion to over 2 trillion parameters, and have been released in various versions, including Llama 2, Llama 3, and Llama 4. Initially a foundation model, later versions included instruction fine-tuned variants, and the models have been made available under licenses that permit some commercial use, though not fully open source.
The release of LLaMA marked a significant step in the democratization of generative AI, as it provided a powerful alternative to proprietary models like OpenAI's GPT series. Its open distribution, especially after the unauthorized leak of the first version, spurred a wave of research and development in the AI community.
Background
Following the release of models like GPT-3, the AI research community focused on scaling up models, which led to emergent capabilities. The launch of ChatGPT in late 2022 intensified interest in large language models. Meta's Chief AI scientist, Yann LeCun, noted that large language models are particularly useful for aiding with writing tasks, positioning LLaMA as a tool for such applications.
Initial Release and Leak
The first version of LLaMA was announced on February 24, 2023, via a blog post and a paper detailing its training, architecture, and performance. The inference code was released under the open-source GPLv3 license, but access to the model weights was restricted to academic researchers and industry labs on a case-by-case basis. Meta trained models of various sizes, from 7B to 65B parameters, on publicly available data, aiming to make them accessible on different hardware.
On March 3, 2023, a torrent with LLaMA's weights was uploaded to 4chan, and the link spread through online AI communities. This unauthorized leak led to widespread availability, with Meta filing takedown requests against HuggingFace repositories and a GitHub script. Reactions were mixed: some feared misuse, while others celebrated the accessibility, comparing it to the open release of Stable Diffusion, which accelerated innovation.
Llama 2
On July 18, 2023, Meta announced Llama 2 in partnership with Microsoft, releasing models with 7, 13, and 70 billion parameters. The architecture was largely unchanged from Llama 1, but training data was increased by 40%. Llama 2 included both foundation and chat fine-tuned models, and weights were released for many commercial uses, though the license's acceptable use policy meant it was not fully open source. Code Llama, a fine-tune for code, was released in August 2023 and January 2024.
Llama 3 and Beyond
On April 18, 2024, Meta released Llama 3 with 8B and 70B parameter sizes, pre-trained on approximately 15 trillion tokens. The models showed strong performance, beating Gemini Pro 1.5 and Claude 3 Sonnet on many benchmarks. Meta also announced plans for multilingual and multimodal capabilities. Llama 3.1 followed on July 23, 2024, and Llama 4 was released in April 2025. In April 2026, Meta Superintelligence Labs introduced Muse Spark as a replacement for Llama.
Impact and Reception
The LLaMA family has had a profound impact on the AI ecosystem, enabling researchers and developers to build on a powerful open model. Its release has sparked debates about open-source licensing and the ethics of model distribution, while also fostering a vibrant community of tools and applications.