Meta unveiled Llama 3, the latest iteration of its large language model (LLM), on April 18, hailing it as a significant advancement over Llama 2. This version boasts 8B and 70B parameters initially, with forthcoming models expected to reach 400B parameters. Llama 3 was trained on an expansive dataset of over 15 trillion tokens, surpassing Llama 2 by 7 times in size and containing 4 times more code. It incorporates filtering techniques, including NSFW filters, to enhance data quality.

Llama 3 surpasses Llama 2 and rival models like Anthropic’s Claude Sonnet, Mistral Medium, and OpenAI’s Chat GPT-3.5 in more than half of 12 specified use cases. While the initial releases are text-based, future versions will be multilingual and multimodal, with improved contextual understanding and enhanced capabilities in reasoning and coding.

Meta plans to deploy Llama 3 widely across cloud providers, model API providers, and other platforms. Additionally, Meta has introduced new avenues for end users to access AI services through Meta AI, a dedicated website offering writing assistance, games, simulated job interviews, and homework aid.

Integration with Meta's platforms like Facebook, Instagram, WhatsApp, and Messenger is underway, with plans to extend services to Ray-Ban Meta smart glasses and Meta Quest VR headset. CEO Mark Zuckerberg touted Meta AI as the most advanced AI assistant available for free use.

This announcement follows recent upgrades to competing services, including ChatGPT's transition to GPT-4 Turbo and Microsoft Copilot's adoption of GPT-4 Turbo, intensifying competition in consumer-oriented AI services.