Meta Launches Llama 3.1 Models Including Powerful 405B Variant
Meta unveils its Llama 3.1 collection of open source AI models, featuring a massive 405B parameter variant that rivals top proprietary systems. The new models offer expanded context length, multilingual support, and are immediately available through IBM watsonx.ai.
Meta unveils the Llama 3.1 collection of multilingual large language models, introducing 8B, 70B, and a first-ever 405B parameter variant. The instruction-tuned 405B model stands as the largest and most powerful open source language model currently available, offering performance that rivals the best proprietary models on the market. Users access this cutting-edge technology immediately through IBM watsonx.ai, where they deploy it on the IBM cloud, in hybrid environments, or on-premises.
This new release shows significant progress since the April launch of Llama 3, delivering on Meta's promise to enhance their AI models. The Llama 3.1 family features dramatically increased context length, expanded tool use capabilities, and broad multilingual support. These upgrades give developers a more versatile and capable open source foundation for building complex AI applications.
The launch also advances the goals of the AI Alliance, a collaborative effort founded by Meta and IBM in late 2023. With over 100 members spanning industry, academia, and government, the Alliance promotes accessible, safe, and responsible AI innovation. By releasing Llama 3.1 as an open, state-of-the-art model ecosystem, Meta provides the global AI community with essential tools to experiment, build, and responsibly scale new ideas.