The open-source new king Llama 3.1 405B has officially launched, surpassing closed-source SOTA models like GPT-4o and Claude 3.5 Sonnet in multiple benchmark tests.
Key points of the Llama 3.1 model family:
- 405B model performance comparable to the best closed-source models
- Open-source/free-to-use weights and code, allowing fine-tuning, distillation, and deployment
- 128k context, multilingual, good code generation and complex reasoning abilities
- Easy integration with Llama Stack API
Meta also released a detailed 90+ page technical paper covering various aspects of large model research.
Meta CEO Zuckerberg emphasized the importance of open-source AI in a post and relaxed licensing terms, allowing developers to use Llama 3.1's output to improve third-party AI models.
Industry insiders generally consider this a significant milestone in AI development:
- Karpathy called it the first time GPT-4 level cutting-edge large models are open to everyone
- Yann LeCun stated that open-source LLMs are now on par with or even better than closed-source LLMs
- Jim Fan called it a historic moment, with GPT-4's power now in our hands
The release of Llama 3.1 marks a new era for open-source AI models, potentially accelerating the development of the entire AI ecosystem.