Overview
Llama 3 is Meta’s strongest open-source LLM, available in 8B and 70B versions, substantially outperforming its predecessor in reasoning, code generation, and instruction following.
Key Capabilities
- Open & Commercial: MIT license, free for research and commercial use
- Long Context: 8K token context support
- Top-tier Code Generation: Outperforms GPT-3.5 on HumanEval
- Multilingual: Trained on 30+ languages
Impact
Llama 3 pushed the boundary of open-source LLMs, becoming one of the most important base models in the AI open-source community.