Perplexity API (Sonar models): pricing, free tier, rate limits and how to start

Choose the Perplexity API product that fits the response you need, then connect it to the correct billing and request contract. This hub covers the tracked Sonar catalog and the current Agent API migration path.

What Perplexity offers through the API

Perplexity’s current platform includes Router, Agent, Search and Embeddings products, while Sonar appears in its legacy API documentation. Agent supports researched responses, Search returns ranked results, Router serves model requests and Embeddings supports vector workflows. Select the product before copying an endpoint or response parser. Official Perplexity documentation.

The official Sonar documentation states that Sonar support continues until September 27, 2026 and directs developers to the Agent API migration guide. Official Perplexity documentation.

This product distinction changes an integration’s architecture. If you need a researched answer with source references, decide where evidence review belongs in the application. If you need ranked documents for your own processing, preserve the search result structure rather than discarding it into an answer string. If you need a direct model completion, use that product’s documented request and rate contract. A shared account key does not make these outputs interchangeable.

For an existing Sonar application, start migration from the current behavior that users depend on. Save a safe prompt, accepted answer structure, citation handling and completion state. The official migration reference maps messages, output handling, search controls and asynchronous behavior to the Agent API. Treat those as separate checks instead of only replacing a URL. Official Perplexity documentation.

Evaluate evidence quality as part of the task. Open the returned sources, check whether they support the nearby claims and distinguish original documentation from commentary. Include a question where the requested information is not available. The application should be able to communicate that limitation instead of treating every fluent paragraph as a researched result. Your acceptance criterion should describe both answer quality and evidence handling.

The official native Python and TypeScript SDKs expose the current platform, and compatible interfaces are documented for existing client ecosystems. Keep the selected method and its response type visible in your integration record. Sonar code that reads a completion choice is different from Agent code that reads response output. Preserve the appropriate citations and metadata alongside text. Official Perplexity documentation. Official Perplexity documentation.

For a research assistant, design a fixture with a clear question and source preference. For a documentation tool, include a version-sensitive question so stale evidence is noticeable. For a summarizer, decide whether it should summarize supplied material or search for additional context. Those are different tasks and should be evaluated separately, even if the interface presents them through the same text box.

Maintain the product choice, request configuration, safe fixture and acceptance criteria together. Add the official reference for each optional feature you enable. That record makes a later model, preset or endpoint change reviewable. It also helps explain why two integrations that both use a Perplexity key can have different usage categories and failure handling.

Pricing at a glance

The tracked live catalog below is Sonar-focused. Read Perplexity pricing by product before applying it to Router, Agent or Search. Those products have separate billing structures. Use the AI API cost calculator for supported model-token assumptions and retain any additional request or tool charges separately; do not treat a Sonar row as the price of an Agent preset.

Models available

Open the Sonar model catalog and migration guide for documented identifiers and replacement planning. The live table is not a complete inventory of every model exposed by Perplexity’s other products. Follow the relevant official catalog when selecting a Router model or an Agent configuration, and retain the endpoint used with that choice.

Before adding a selector to a product UI, decide what each option promises the user. A researched answer, ranked search results and a direct completion are different outputs. Give them clear labels and preserve their source and completion metadata. That design prevents an endpoint change from silently altering the artifact a user thinks they requested.

Free tier and how to get a key

Use Perplexity free-access conditions to distinguish purchased API credit, an explicit promotion and a consumer subscription. Follow project and API key setup to establish the account that will own the workload. Confirm its usable credit and payment preferences before launching an unattended research job.

Rate limits and tiers

Perplexity publishes product-specific admission rules and Sonar model limits by usage tier. The current tier reference bases progression on cumulative credit purchases rather than current balance. Read Perplexity request and tier limits before applying a single rate setting to every product. Official Perplexity documentation.

Keep an application queue with a clear state for waiting and active work. Research questions can remain useful for different lengths of time, so record a deadline and avoid executing stale requests after a long backlog. Separate retry attempts from new user tasks so monitoring can explain what the application actually dispatched.

Service status

Use the official incident page for service context when several unrelated requests fail. Your account can still encounter authentication, billing or validation problems while the public status is healthy. Follow Perplexity error diagnosis using the exact product, safe error detail and response status. Do not replace a missing response with an answer-shaped success message.

Start in 5 minutes

from perplexity import Perplexity

client = Perplexity()
response = client.responses.create(
    preset="low",
    input="Explain HTTP conditional requests using primary documentation. "
          "Keep the answer brief and include source references.",
)
print(response.output_text)

This current Agent API example uses the native client and prints its text output. Generated wording and sources vary. The complete Perplexity Python tutorial adds local credential validation, safe error handling and inspection of the returned output structure. Official Perplexity documentation.

Run a harmless question first and inspect its cited evidence before attaching the request to a public UI. Then test a representative task and save its safe configuration. Treat the first successful call as the start of evaluation: it establishes connectivity and response handling, while your acceptance cases establish whether the output is useful.

Recent changes

Review the official migration notice together with observed data changes. A change in product contract can require more work than a rate update. Keep the old Sonar fixture as a regression test while adapting request fields, citations and asynchronous state to Agent. Publish the application change only after those behaviors have been checked together.

Frequently asked questions

Sources

• Current platform quickstart ↗
• Sonar API ↗
• Migration and support notice ↗
• Product pricing ↗
• Rate and tier rules ↗

Last verified 12 Sep 2026 · Source ↗