Meta Releases Llama 3.1 405B, Its Biggest Open-Source AI Model

llama 3.1 announced by meta
Image Courtesy: Meta
In Short
  • Meta has unveiled its largest AI model, Llama 3.1 405B, and open-sourced it. It rivals OpenAI's GPT-4o model and even beats it in key benchmarks.
  • Apart from that, Meta has also released an upgraded Llama 3.1 70B and 8B models.
  • All three models have a context window of 128K tokens and support eight languages. Multimodal support is not enabled yet.

After a long wait, Meta has finally released Llama 3.1 405B, its largest AI model yet, trained on 405 billion parameters. Not only that, Meta has also released an upgraded family of models which include Llama 3.1 70B and Llama 3.1 8B models. And all of these models are open-source and the Zuckerberg-led company says that “Meta is committed to openly accessible AI.

All three Llama 3.1 models come with a context length of 128K tokens and support eight different languages. So Meta has incorporated a large context window and multilingual capability. As for benchmarks, the largest Llama 3.1 405B model outranks OpenAI’s leading AI models such as GPT-4 and the latest, GPT-4o (Omni).

In the MMLU benchmark, Llama 3.1 405B scores 88.6 points whereas GPT-4o gets 88.7 points, meaning both are almost on the same level. Apart from this, in nearly all other tests including MBPP, GSM8K, ARC Challenge, etc., the 405B model outcompetes GPT-4o. The only major benchmark in which Llama 3.1 405B trails is HumanEval, but by a small margin.

Image Courtesy: Meta

In HumanEval, GPT-4o scores 90.2 whereas the 405B model scores 89.0 points. Finally, coming to multimodality, sadly, Meta has not released a truly multimodal model yet, even after the Llama 3.1 release. Meta says Llama 3.1 models support image, video, and speech recognition, however, they are still under active development and not yet ready for release.

If you want to check out how the Llama 3.1 405B model performs, you can head over to groq.com and chat with the frontier model. Keep in mind, due to high traffic, it’s facing some server issues at the moment. Apart from that, if you are in the US, you can access the 405B model on WhatsApp, and meta.ai (visit).

SOURCE Meta
Comments 0
Leave a Reply