Meta rolls out its biggest open source AI model Llama 3.1 405B

Tech firm Meta has unveiled its largest open-source AI model, Llama 3.1 405B.

The model boasts 405 billion parameters, making it one of the most advanced and capable models in the open-source AI space, said the company.

The model will be available free of charge, enabling widespread access to the developer community. It was trained using 16,000 NVIDIA's H100 GPU (graphic processing unit) and leverages advanced training and development techniques.

Meta claims the model can outperform OpenAI's GPT-4o and Anthropic’s Claude 3.5 Sonnet on several benchmarks. Developers can run inference with Llama 3.1 405B on their own infrastructure at approximately half the cost of using other proprietary models such as GPT-4o—for both user-facing and offline inference applications, it said.

Towards an open and positive AI future

In an open letter, Meta CEO Mark Zuckerberg emphasised the importance of open source for a positive AI future. 

He stated that open-source AI models are not only catching up but are already advancing faster than proprietary models, akin to how Linux evolved to become industry standard foundation, for both cloud computing and the operating systems that run most mobile devices. 

“I believe that open source is necessary for a positive AI future. AI has more potential than any other modern technology to increase human productivity, creativity, and quality of life–and to accelerate economic growth while unlocking progress in medical and scientific research.

"Open source will ensure that more people around the world have access to the benefits and opportunities of AI, that power isn’t concentrated in the hands of a small number of companies, and that the technology can be deployed more evenly and safely across society,” said Zuckerberg. 

The company is actively building partnerships with companies such as Scale.AI, Dell, and Deloitte, which aim to assist enterprises in adopting Llama and train custom models using their own data.

Llama 3.1 405B is available on cloud platforms such as AWS, Azure, and Google Cloud.

“With past Llama models, Meta developed them for ourselves and then released them, but didn’t focus much on building a broader ecosystem. We’re taking a different approach with this release. We’re building teams internally to enable as many developers and partners as possible to use Llama, and we’re actively building partnerships so that more companies in the ecosystem can offer unique functionality to their customers as well,” said Zuckerberg.

He further stated that Meta’s safety process involves rigorous testing and red teaming to identify potential harms from their models, aiming to mitigate risks before release. Since the models are open-source, external parties can also conduct their own tests. 

Zuckerburg also drew a comparison between Meta’s investment in open-source AI and its previous Open Compute Project and claimed to have saved the company "billions".

"Meta has a long history of open-source projects and successes. We’ve saved billions of dollars by releasing our server, network, and data center designs with Open Compute Project and having supply chains standardize on our designs. This approach has consistently worked for us when we stick with it over the long term," he said.