From the archive
Llama 3.1 adds a 405B model and longer multilingual context
Meta releases Llama 3.1 models with 128,000-token context, eight-language support, and revised output-use terms.
Meta released Llama 3.1 on July 23, 2024. The family introduced a 405-billion parameter model alongside updated 8-billion and 70-billion parameter versions.
More context and language support
The models offered a 128,000-token context window and support for eight languages. That expanded the material available for long-document processing, coding, and multilingual conversations.
Meta also changed the license to permit Llama outputs to improve other models. This supported uses such as synthetic training data and distillation, subject to the license terms.
Weights and deployment tools
The models were released for download through Meta and Hugging Face , with hosting partners providing other access routes. Meta included Llama Guard 3 and Prompt Guard and opened a request for comment on the Llama Stack interfaces.
The release combined larger model weights with tools for assembling applications around them, rather than supplying only a single hosted chat endpoint.