Llama 3
Llama 3 is Meta's open-source next-generation large language model, available in 8B and 70B sizes and trained on over 15 trillion tokens. It leads its class on reasoning, coding, and instruction-following benchmarks.
Verification level not recorded · · Submit a correction
Best for developers and researchers building their own AI applications; not suitable for non-technical users wanting a ready-made product.
Decision facts
“Not verified” means evidence is insufficient, not that the capability is absent.
What is Llama 3
Llama 3 is Meta's open-source next-generation large language model, available in 8B and 70B sizes and trained on over 15 trillion tokens. It leads its class on reasoning, coding, and instruction-following benchmarks.
Key features of Llama 3
- Self-hosting an open-source LLM
- Fine-tuning domain-specific assistants
- Code generation and programming help
- Academic research and model evaluation
Good for
- Open source and free to deploy
- Two sizes (8B and 70B) for different needs
- 7x more training data than the previous generation
Watch out
- Self-hosting requires significant compute
- Fine-tuning and safety alignment are your responsibility
- No turnkey consumer product included
How to use Llama 3
- Visit llama.meta.com/llama3
- Get the weights from Hugging Face or GitHub
- Review and accept Meta's license
- Set up an inference environment and load the model
- Fine-tune and integrate into your application
Who Llama 3 is for
Difficulty: Advanced
- Self-hosting an open-source LLM
- Fine-tuning domain-specific assistants
- Code generation and programming help
- Academic research and model evaluation
FAQ
Is Llama 3 free?
Yes, it is open source and free to download under Meta's license terms.
What model sizes are available?
8B and 70B versions are available now, with a 400B model announced as in training.
How good is its Chinese?
Pretraining data covers 30+ non-English languages; evaluate Chinese quality for your use case.
Sources and verification
Evidence status: Verification level not recorded
Sources: llama.meta.com (opens in a new tab)
Content reviewed: · Submit a correction →