Llama 3.3

Llama 3.3 (70B) is Meta's multilingual LLM, delivering state-of-the-art performance comparable to the Llama 3.1 405B model while optimizing for hardware efficiency.

This is the Meta Llama 3.3 instruction-tuned model, released December 6, 2024. The 70B parameter version is a major step: it matches the performance of the much larger Llama 3.1 405B model, but with significantly reduced computational demands. We trained it on over 15 trillion tokens, expanded the context window to 128k, and optimized it for multilingual dialogue (supporting eight languages including English, Spanish, and Hindi). It's designed for commercial and research use, offering high-end LLM capability with practical, accessible deployment.

https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct