New Perplexity Pro Sonar Huge model leverages LLAMA 3.1 405b for AI Search

Perplexity has quietly introduced the Sonar Huge model for Perplexity Pro users. The Sonar models, post-trained by the Perplexity team, are specifically optimized for search tasks. Previously, the Sonar large model was based on Meta's LLAMA 3.1 70b model. The new Sonar Huge model, however, is built on the latest LLAMA 3.1 405b model. This is a significant development, as it's the first time a Sonar model designed for search is based on such a powerful frontier model. Early results are highly promising, and it will be interesting to see how this performs in real-world applications.

In addition to this, there's a subtle but impactful change: Perplexity can now search across X (formerly Twitter) in addition to Reddit.

This enhancement greatly improves Perplexity's ability to gather information on the most recent events, making the Perplexity Pro subscription even more valuable!

Another notable update is the addition of a bookmark button on the Perplexity web Discover page. This feature allows users to save pages to their bookmarks collection without having to open them, making it easier to read them later. These updates are quite exciting, and although they haven't been officially announced yet, it's likely that the Perplexity team will release their evaluation results soon.