Perplexity explained: search, AI agents and a business built across models

Perplexity sells a way to find information and get work done across several AI models. Its products can combine models from OpenAI, Anthropic, Google and other developers with Perplexity’s search, prompts, tools and usage rules. Choosing a Claude model inside Perplexity therefore gives you Claude operating within Perplexity’s service, with a different surrounding experience from Anthropic’s own app. Its published Search model catalogue also includes Perplexity’s own Sonar 2.[2]

Perplexity was founded in 2022. Co-founder Aravind Srinivas brought a research background at Google, DeepMind and OpenAI; he discussed the move from researcher to entrepreneur at UC Berkeley, where he earned his computer-science PhD.[1]

The company’s distinction is the system it assembles around the models. Search finds and synthesises information. Computer carries out longer tasks with tools and connected applications. The Comet browser provides another place to ask questions and delegate browser work. These are different routes into Perplexity, with their own controls and entitlements.[3][2]

What Perplexity builds itself

Perplexity operates a search service that can return ranked web results, extracted text and controls for domains, languages and regions. Developers can buy that retrieval service through an application programming interface (API), which lets their software request results directly. A company building its own assistant can therefore use Perplexity to find material while deciding for itself how to process the results.[6]

Perplexity’s research programme covers retrieval, ranking, agent training and serving systems. September publications describe Photon, a retrieval engine, and work to improve a Computer model’s behaviour and reduce tool errors. Together, these cover both the algorithms handling information and the infrastructure needed to serve users.[8]

The Agent API makes the commercial logic particularly visible. One interface gives a software builder access to several model providers, with consistent controls for reasoning, tokens and tools. Perplexity supplies that integration; the underlying intelligence can come from another company. The buyer is paying for a useful combination of services, even when the model name is familiar.[7]

How the company charges

Perplexity has free and paid individual plans, business plans and separately billed APIs. Enterprise products add organisation management and internal knowledge search through shared Projects and centrally managed file connectors. That gives it a second buyer alongside the individual researcher: an employer that wants the service available across a team.[4]

The distinction affects everyday use. Perplexity’s credit guide, updated October 6, says Ask searches do not consume Computer credits. Paid Consumer Pro accounts are gradually receiving weekly Computer usage; eligibility appears under Included in Pro. Consumer Max includes 10,000 credits monthly. At the documented rate of 100 credits per US$1, that allocation represents US$100 of credit value, not a fixed number of completed projects.[5]

A long task can consume more than a short question because Computer may research several items, use tools and produce files. This makes the value of the service partly organisational: preserving context, choosing the next step and handing work between capabilities. The amount of work requested changes what must happen behind the interface.[3]

Its services can appear inside another company’s product

Samsung’s Galaxy S26 is one example. Samsung’s UK launch material identifies Perplexity as a pre-installed assistant, requiring a network connection and Perplexity account. Perplexity’s own announcement also says its APIs supply search and reasoning to Samsung’s Bixby. The visible assistant can therefore belong to Samsung while a service underneath it comes from Perplexity.[9][10]

That creates two routes to customers: persuade people to open Perplexity itself, or supply capabilities where they already spend time. An integration changes how customers encounter the company’s technology. It does not establish how many people become active users or paying subscribers.[9][10]

The company’s position is easiest to understand through that combination. Perplexity builds search and agent systems, incorporates models from other developers, and sells the result through applications, subscriptions and APIs. Its commercial task is to make the combined service useful enough that customers keep choosing it as the models underneath continue to change.[7][4]

Report an error or suggest a correction