In March 2023, the model weights for Meta's first LLaMA large language model were leaked online, making the weights publicly accessible via BitTorrent and other platforms. The leak, which occurred shortly after the model's initial restricted release to researchers, allowed a broad community of developers and researchers to download, run, and fine-tune the model, significantly accelerating open-source AI development and sparking debates about accessibility, safety, and the future of AI distribution.
Background
Following the release of large language models such as GPT-3, a major focus of AI research was up-scaling models, which in some instances showed major increases in emergent capabilities. The release of ChatGPT and its surprise success caused a surge in attention to large language models, prompting many organizations to accelerate their own AI programs.
Compared with other responses to ChatGPT, Meta's Chief AI scientist Yann LeCun stated that large language models are best for aiding with writing. Meta's approach to releasing its LLaMA model differed from that of OpenAI, which kept GPT-3 and ChatGPT weights proprietary.
Initial release of LLaMA
The first version of LLaMA (stylized as LLaMA and sometimes referred to as Llama 1) was announced on February 24, 2023, via a blog post and a paper describing the model's training, architecture, and performance. The inference code used to run the model was publicly released under the open-source GPLv3 license. Access to the model's weights was managed by an application process, with access to be granted "on a case-by-case basis to academic researchers; those affiliated with organizations in government, civil society, and academia; and industry research laboratories around the world".
LLaMA was trained on only publicly available information and was trained at various model sizes, with the intention to make it more accessible to different hardware. The model was exclusively a foundation model, although the paper contained examples of instruction fine-tuned versions of the model.
Meta AI reported that the 13B parameter model's performance on most NLP benchmarks exceeded that of the much larger GPT-3 (with 175B parameters), and the largest 65B model was competitive with state-of-the-art models such as PaLM and Chinchilla.
The leak
On March 3, 2023, a torrent containing LLaMA's weights was uploaded, with a link to the torrent shared on the 4chan imageboard and subsequently spread through online AI communities. That same day, a pull request on the main LLaMA repository was opened, requesting to add the magnet link to the official documentation. On March 4, a pull request was opened to add links to HuggingFace repositories containing the model.
On March 6, Meta filed takedown requests to remove the HuggingFace repositories linked in the pull request, characterizing it as "unauthorized distribution" of the model. HuggingFace complied with the requests. On March 20, Meta filed a DMCA takedown request for copyright infringement against a repository containing a script that downloaded LLaMA from a mirror, and GitHub complied the next day.
Reactions to the leak
Reactions to the leak varied. Some speculated that the model would be used for malicious purposes, such as more sophisticated spam. Others celebrated the model's accessibility, as well as the fact that smaller versions of the model can be run relatively cheaply, suggesting that this would promote the flourishing of additional research developments.
Multiple commentators, such as Simon Willison, compared LLaMA to Stable Diffusion, a text-to-image model which, unlike comparably sophisticated models that preceded it, was openly distributed, leading to a rapid proliferation of associated tools, techniques, and software. The leak effectively turned LLaMA into an open-weight model, bypassing Meta's controlled access and enabling a wave of community-driven experimentation.
Impact on open-source AI
The leak accelerated the development of the open-source AI ecosystem. Within weeks, developers created fine-tuned versions of LLaMA for specific tasks, including instruction-following and chat, using techniques like LoRA and QLoRA to run the model on consumer hardware. This led to a proliferation of tools and applications, including local chatbots and research projects, that would not have been possible under the restricted access.
The leak also influenced Meta's subsequent release strategy. The success and community enthusiasm around the leaked model likely contributed to Meta's decision to release Llama 2 in July 2023 with a more permissive license that allowed commercial use, albeit with restrictions. This marked a shift from the initial case-by-case academic access to a broader public release, though the license still prohibited certain uses, leading to disputes over whether it qualified as open source.
Subsequent versions
Following the leak, Meta continued to develop the LLaMA family. On July 18, 2023, in partnership with Microsoft, Meta announced Llama 2, the next generation of Llama. Meta trained and released Llama 2 in three model sizes: 7, 13, and 70 billion parameters. The model architecture remained largely unchanged from that of Llama 1, but 40% more data was used to train the foundational models.
Llama 2 included foundation models and models fine-tuned for chat. In a further departure from the original version, all models were released with weights and could be used for many commercial use cases. Because Llama's license enforced an acceptable use policy that prohibited certain purposes, it was not open source in the strict sense. Meta's use of the term "open-source" to describe Llama was disputed by the Open Source Initiative (which maintains The Open Source Definition) and others.
Code Llama, a fine-tune of Llama 2 with code-specific datasets, was released in versions of 7B, 13B, and 34B parameters on August 24, 2023, with a 70B version released on January 29, 2024. Starting with the foundation models from Llama 2, Meta AI trained an additional 500B tokens of code datasets, followed by 20B tokens of long-context data, creating the Code Llama foundation models. This foundation model was further trained on 5B instruction-following tokens to create the instruct fine-tune. Another foundation model was created for Python code, trained on 100B tokens of Python-only code, before the long-context data.
On April 18, 2024, Meta released Llama 3 with two sizes: 8B and 70B parameters. The models were pre-trained on approximately 15 trillion tokens of text gathered from "publicly available sources", with the instruct models fine-tuned on "publicly available instruction datasets, as well as over 10M human-annotated examples". Meta AI's testing showed in April 2024 that Llama 3 70B was beating Gemini Pro 1.5 and Claude 3 Sonnet on most benchmarks. Meta also announced plans to make Llama 3 multilingual and multimodal, better at coding and reasoning, and to increase its context window.
Regarding scaling laws, Llama 3 models empirically showed that when a model is trained on data that is more than the "Chinchilla-optimal" amount, performance continues to scale log-linearly. For example, the Chinchilla-optimal dataset for Llama 3 8B is 200 billion tokens, but performance continued to scale log-linearly to the 75-times larger dataset of 15 trillion tokens.
During an interview with Dwarkesh Patel, Mark Zuckerberg said that the 8B version of Llama 3 was nearly as powerful as the largest Llama 2. Compared to previous models, Zuckerberg stated the team was surprised that the 70B model was still learning even at the end of the 15T tokens training. The decision was made to end training to focus GPU power elsewhere.
Alongside the release of Llama 3, Meta rolled out Meta AI, an AI assistant built on Llama. Meta AI has a dedicated website and is available on Facebook and WhatsApp. The latest version is Llama 4, released in April 2025. In April 2026, Meta Superintelligence Labs released Muse Spark as a replacement for Llama powering chatbots on various Meta services.
Legacy
The LLaMA leak of 2023 is widely regarded as a pivotal moment in the history of open-source AI. It demonstrated the demand for accessible large language models and influenced the strategies of both Meta and other organizations. The leak also highlighted the challenges of controlling the distribution of powerful AI models, a theme that continues to shape policy discussions around AI safety and governance.
The event contributed to the broader generative AI boom, as the availability of LLaMA weights enabled a wave of innovation that might otherwise have been delayed. It also set a precedent for the release of subsequent models like Llama 2 and Llama 3, which have become foundational tools for researchers and developers worldwide.