Meta launches its biggest AI model - Llama 3.1, and it's free and open-source: What it means for users?

Meta has launched its next generation of Large Language Model (LLM) dubbed Llama 3.1. It’s the first openly available AI model that vies against the existing closed-system AI models such as GPT-4/GPT4–Omni and Claude 3.5 Sonnet. Meta claims that it’s the biggest open-source model, and it’s smarter and better at human evaluation. It has more advanced reasoning capabilities, making it solve complex queries that users throw at it.

Meta’s Llama 3.1 is free and open-source

Meta says that the Llama 3.1 is the most advanced open-source model. It beats GPT-4, GPT4-Omni, and Claude 3.5 Sonnet in several departments. It is an upgrade over the Llama 3 as it has more complex query-solving capabilities. The credit for this goes to the amount of training it has gone through.

The largest version of the model has 405 billion (405B) parameters and was trained with more than 16,000 Nvidia H100 GPUs. Despite rigorous training and spending roughly millions of dollars, Meta is giving Llama 3.1’s 405-billion parameter model for free. However, it will be free for some time and then will switch users back to the 70-billion (70B) version. That’s likely because it’s costly.

“As our largest model yet, training Llama 3.1 405B on over 15 trillion tokens was a major challenge. To enable training runs at this scale and achieve the results we have in a reasonable amount of time, we significantly optimized our full training stack and pushed our model training to over 16 thousand H100 GPUs, making the 405B the first Llama model trained at this scale,” noted Meta in its blog post.

Alongside the latest 405B version of the model, the 8B and 70B versions have also been upgraded. These versions are multilingual and have a longer context length of 128K for better reasoning capabilities.

Meta has also made changes to its license, as now other developers can use outputs from Llama models, including the 405B version, to improve other models. Since it’s open-source, developers can download the models on Meta’s Llama website.

What this means for users?

Keeping all the numbers aside, what does this new Llama 3.1 model mean for the general users? It simply means that AI in Meta apps will be smarter than ever and you can rely more on it. In fact, Meta CEO Mark Zuckerberg has already demonstrated the power of the model on Instagram. The latest model is used in Meta AI, the company’s AI tool/chatbot that lets you ask queries and generate images. The latest feature of Meta AI is “Image Me.” The feature takes a snapshot of your face and uses the data to generate images for you. This means you can say “Image me as a gladiator,” just like Zuckerberg did, and you can see a picture of you as a gladiator. Now, the accuracy of the tool has improved a lot as seen in the demonstration. The feature powered by Llama creates images that are real-life-like.

Meta AI is available on most Meta apps like Instagram, Facebook,  and WhatsApp. It’s now also available in local languages in different such as Hindi, French, German, Spanish, and Italian.

Since Llama 3.1 is openly available, developers can use it for their apps and services. That said, this will not only benefit users of Meta apps, but it will possibly make the competition stiffer. Consequently, benefiting users with varied and more advanced AI options available.