According to TechCrunch, Nick Clegg, Meta's Global Affairs President, stated at an event in London on Tuesday, March 10th, "We hope to launch our next-generation foundational model suite, Llama 3, in the next month, possibly even sooner, in a very short period of time." "We will release a series of models with different functionalities and purposes within this year, and it will start very soon," he added. Initially reported by The Information, a smaller version of Llama 3 may be released as early as next week, with the complete open-source model scheduled for release in July. This version will be able to compete with Claude 3 and GPT-4. Chris Cox, Meta's Chief Product Officer, further stated that the plan is to leverage Llama 3 to power various Meta products. This release comes as Meta attempts to catch up in the competitive field of generative AI. Its predecessor, Llama 2, was released in July 2023 and received criticism due to its limitations. Joelle Pineau, Vice President of AI Research, said, "Our goal is to develop Meta AI powered by Llama into the most useful assistant in the world. However, achieving this goal requires a lot of work." What is Llama 3? Llama 3 is a large-scale language model that ranges in size from very small versions designed to compete with models like Claude Haiku or Gemini Nano, to larger versions that are fully equipped for response and inference, similar to GPT-4 or Claude Opus. However, detailed information about Llama 3 is still scarce. It is expected to follow the footsteps of its previous version, being open-source and likely to support multimodality, enabling simultaneous understanding of visual and textual inputs. It is speculated that Llama 3 will be released in different versions and sizes, with the smallest version having 7 billion parameters and the largest version approximately 140 billion parameters. However, this is still far fewer parameters than those used to train GPT-4.