Unlock the Secrets of Sonar: Perplexity’s Revolutionary Answer Engine Explained

For several months now, generative AI has been making significant inroads into the realm of search technology. In this space, the American startup Perplexity has quietly advanced its chess pieces with a tool named Sonar. Its aim is to offer a viable alternative to Google by blending speed, accuracy, and reliable sources. The company’s strategy relies on a combination of technological performance and partnerships with media outlets like Le Monde, Time Magazine, and Der Spiegel.

A Real-Time Web-Connected Response Engine

Launched in 2024, Sonar is an AI-based response engine. Unlike traditional LLMs, Sonar does not merely “hallucinate” responses from a static dataset. Instead, it queries the web in real-time to answer each question, then generates a structured synthesis with links to its sources. This blend of processing speed, readability, and transparency makes it one of the most sought-after AI models in document research today.

Under the hood, Perplexity utilizes Meta’s open-source model Llama 3.3 (70B), which the startup has internally refined to enhance factuality and readability. According to the company, Sonar processes up to 1,200 tokens per second and has outperformed user-tested models like Claude 3.5 Haiku or GPT-4o mini. It is available in two versions:

• A quick and lightweight Sonar version,
• A more powerful Sonar Pro version, capable of handling simultaneous queries and customizable citations.

Zoom, a Quintessential Use Case for Sonar

This technology is not limited to Perplexity’s interface alone. Zoom, the video conferencing tool, has integrated Sonar into its AI Companion 2.0. This allows meeting participants to ask questions and receive immediate, sourced, and contextualized answers without leaving their video conferencing window.

This type of integration exemplifies Sonar’s positioning: an AI component designed to fit seamlessly into existing tools, offering quick, reliable, and actionable responses. Targeting sectors like healthcare, education, and digital services, Perplexity focuses on areas where information retrieval needs to be both swift and trustworthy.

Our goal is to make extended linguistic models based on research accessible and cost-effective, stated Aravind Srinivas, CEO of Perplexity.

In the same vein, developers and businesses can incorporate real-time AI-powered search within their various products and services to enhance user experience with presumably reliable and updated information.

With Le Monde, a New Form of Editorial Engine

But Perplexity’s strategy extends beyond typical public or B2B use. In May 2025, the national daily Le Monde formalized a multi-year agreement with the American startup. This partnership will integrate Sonar into the newspaper’s interfaces (website and applications), enabling readers to pose questions in natural language and receive answers based solely on content from Le Monde.

This agreement provides us access to Sonar, a response engine that will be progressively deployed in the coming month on our site and in our apps. This new feature will thus enable Le Monde’s readers to have a research experience that complements what a traditional search engine offers, explains Le Monde.

This partnership signifies a notable shift in the relationship between generative AI and media. On one side, Perplexity benefits from quality content to fuel its response engine. On the other, Le Monde retains its rights (the contents are not used for training models) and ensures payment for each referenced use. The agreement also ensures control over the quality of responses: if they lack rigor, Le Monde reserves the right to suspend the integration.

[This partnership] will enable our title to attract a new generation of readers, already very inclined towards using these technologies.

More than just a technological addition, this partnership reflects a shift in balance. The press, long a spectator or victim of content appropriation by platforms, is now negotiating terms of use with AI players while benefiting from the advantages these various tools and models offer.

A Developing Ecosystem Strategy

With Sonar, Perplexity embodies a new approach to search that takes the form of an immediate, well-argued response built from live sources. However, this concept also raises some questions: how to ensure the reliability of sources in contexts less regulated than press partnerships? What role remains for traditional search engines in a world where people no longer “search” but “ask”? And, more broadly, how far will media companies go in delegating the relationship with their content to AI assistants?

For now, Perplexity is proceeding cautiously. Its agreements are not exclusive, its models remain transparent about their sources, and the company stops short of claiming to completely replace traditional search engines. But the momentum has begun. Major AI platforms are now signing agreements with publishers, rewriting the rules of the game established by Google.