Meta has launched a new version of its open-source large language model (LLM).

The social media company claims that Llama 3, which will power its virtual assistant Meta AI, is the “most capable” LLM available to date.

Meta said that compared to Llama 2, its latest model can encode language more efficiently which improves its performance.

Llama 3 was trained on a dataset four times larger than that of Llama 2 and includes four times more code.

Meta said it aims to make Llama 3 multilingual and multimodal, have longer context, as well as improve its reasoning and coding. To achieve these goals, Llama 3 was trained using non-English data covering around 30 languages.

Llama 3 models will be available on a variety of services including Amazon Web Services, Google Cloud and Microsoft Azure.

Meta said that it has put safety measures in place including filters for content.

It is also offering resources to encourage responsible use such as trust and safety tools.

“We’re committed to the continued growth and development of an open AI ecosystem for releasing our models responsibly,” said Meta. “We have long believed that openness leads to better, safer products, faster innovation, and a healthier overall market.

"We’re taking a community-first approach with Llama 3, and starting today, these models are available on the leading cloud, hosting, and hardware platforms with many more to come.”

Earlier this week, Meta announced that it is launching new educational features for its virtual reality (VR) headsets aimed at those over the age of 13. This includes a range of new apps and features for its Quest headsets to aid learning and the application of new skills.

---