Perplexity Will Retire Sonar API Endpoints on September 27, Forcing a Move

Applications built on Perplexity’s Sonar API have a September 27 cutoff. The company says its Sonar chat-completions endpoint will retire that day, requiring developers to migrate to the Agent API responses endpoint to keep using the service.

The change sends developers from Sonar’s existing chat-completions interface to an API built around grounded web search, multi-step research, code execution, built-in tools and access to multiple models through one interface. A Perplexity staff member confirmed in the company’s API forum that users need to complete the migration before the retirement date.

What the replacement adds

• Grounded web search remains available through the Agent API.
• The API adds multi-step research and code execution.
• It provides built-in tools and access to multiple models through one API.

Perplexity is pairing the required move with its own benchmark claims. In its retirement notice, the company said Agent API more than doubles the best Sonar score on BrowseComp and WideSearch. Those figures are company-stated results, rather than an independent measure of how every Sonar application will perform after migration.

An earlier Perplexity API announcement gave more specific comparisons. The company said its fast Agent API preset scored about twice as high as Sonar on DSQA at the same $0.006 per query. It also said its low preset outperformed Sonar Pro by about seven times on BrowseComp while maintaining $0.03 per query. Perplexity further said its high and xhigh presets surpassed Sonar Deep Research in its comparison.

The migration changes more than a request destination. Perplexity says Agent API presets can use tools including web_search, finance_search and fetch_url, and that it re-tunes those presets as frontier and leading open-source models change. That gives migrated workloads access to an evolving tool and model layer managed by Perplexity.

For teams that used Sonar as a search-grounded API, the decision now is whether to adopt that managed operating model. Perplexity’s benchmarks present the new API as a performance upgrade, but its reported results cover particular presets and tests—not every application’s workload.