Llama 3

Open-Weight Languageby Meta AIReleased April 2024

Llama 3 is Meta's open-weight flagship model family, available in 8B and 70B parameter sizes. It significantly narrows the capability gap between open-weight and closed frontier models on key benchmarks. Released under a permissive license, it has become the de facto foundation for open-source AI development globally.

Context Window
8,192 tokens (base); 128K in fine-tuned variants
License
Meta Llama 3 Community License (commercial use allowed)
  • State-of-the-art open-weight performance
  • Fine-tunable for custom use cases
  • On-premise and air-gapped deployment
  • Code generation and analysis
  • Instruction following and chat
  • Foundation for thousands of community fine-tunes
Free to download and run. Inference costs vary by provider: Together AI, Groq, and Replicate offer API access from $0.59/1M tokens (8B) to $0.90/1M tokens (70B).
MMLU: 82.0% | HumanEval: 81.7% | GSM8K: 93.0% | MATH: 50.4%
Llama 3 is Meta's strategic weapon against OpenAI and Google's API ecosystem dominance. By releasing capable open-weight models, Meta commoditizes the model layer while strengthening its position in the fine-tuning and deployment ecosystem. Primary competition: Mistral for open-weight crown.
Related Models
Llama 2Llama 3 8BCode LlamaMistral 7B