Highlights:

  • The API’s standard edition, Sonar, is designed for quick query responses and cost-effectiveness.
  • Perplexity also provides an advanced API version, Sonar Pro, which excels at handling complex queries.

Recently, Perplexity AI Inc. has launched an API that allows developers to integrate search functionality into their applications.

The API is offered in two versions: Sonar and Sonar Pro, both powered by the company’s Sonar series of large language models to handle search queries. Its launch follows reports that Perplexity secured USD 500 million in funding two months ago, achieving a valuation of USD nine billion.

Established in 2022, Perplexity offers an AI-driven search engine bearing its name. Instead of presenting a list of webpage links, the platform delivers direct answers to user queries in natural language. Each response includes citations to the webpages where the retrieved information originated.

The company monetizes through a premium offering, Perplexity Pro, which provides enhanced features. This includes a built-in code interpreter that allows users to both generate and execute code. Additionally, a feature called Shop like a Pro enables one-click e-commerce purchases directly within the search engine’s interface.

The API launched by Perplexity recently introduces a new avenue for the company to monetize its large language models.

The API’s standard edition, Sonar, is designed for rapid query responses and cost-effectiveness. It leverages a Perplexity-developed large language model, also named Sonar, which features a context window of 127,000 tokens. This allows users to include up to 127,000 tokens of data in a single prompt.

Perplexity provides a more advanced API version, Sonar Pro, which excels at handling complex queries. Powered by the Sonar Pro large language model, it features a 200,000-token context window. Additionally, it includes twice as many citations in its prompt responses compared to the standard Sonar API, aiding users in verifying the accuracy of the retrieved information.

Perplexity has not revealed technical specifics about the LLMs behind its API. However, the company has stated that an earlier model in the same LLM series is a customized version of Llama 3.1 70B.

At launch, Perplexity offers several customization options for API customers. Developers can tailor the data sources the API uses to answer search queries. Additionally, users can adjust the Top P and presence penalty settings of the underlying LLMs to minimize nonsensical output and reduce duplicate content in responses.

The Sonar API is priced at USD Five per 1,000 searches, with an additional charge of USD one for every 750,000 words processed or generated by the LLM. Sonar Pro is priced at USD three per 750,000 words of input, while the same volume of output data costs USD 15.

Perplexity reports that thousands of developers are already utilizing its API in various software projects. The company also mentions that one client, the startup Copy.ai Inc., is using the service to save its sales platform customers several hours of work each week.