Meta has officially launched LLaMA 3, its latest large language model designed to rival OpenAI’s GPT-4 and Google’s Gemini. Released under an open-source license, LLaMA 3 sets a new benchmark for transparency, accessibility, and performance in the generative AI race.
According to Meta’s AI blog, the LLaMA 3 family includes models with 8B and 70B parameters — trained on more than 15 trillion tokens. The models outperform Claude, Mistral, and even earlier versions of GPT-4 on several standard benchmarks.
Bloomberg reports that Meta plans to fully open models with 400B+ parameters by the end of the year, aiming to challenge the dominance of closed systems. This release comes as part of Meta’s long-term strategy to build AI that is open, safe, and scalable — and deeply integrated with its family of products, including Facebook, Instagram, and WhatsApp.
Wired highlights how LLaMA 3 is already integrated into Meta AI Assistant, now available in over a dozen countries and across platforms. The assistant can generate text, analyze content, and even help write code — all while running on infrastructure built entirely by Meta.
Why does it matter? Because Meta is pushing against the tide. While OpenAI and Google restrict usage via APIs and licenses, Meta gives direct model weights, encouraging experimentation and independent innovation.
Developers, researchers, and even startups now have access to a powerful tool — with fewer restrictions and strong performance. LLaMA 3 could become the backbone of the open-source AI movement, reshaping how AI is created and distributed in the years to come.