Your Sonar workload runs better on Agent API

Announcements
,
1

Your Sonar workload runs better on Agent API

Higher accuracy at lower cost per query

Thank you for building your research workflows on Sonar. We wanted to reach out about Agent API, our flagship agentic research endpoint. For research-dominant workloads like yours, it delivers higher accuracy at a lower cost per query, with more control.

We benchmarked Agent API against Sonar on DSQA and BrowseComp

  • ~2x accuracy for the same cost: on DSQA, Agent API’s fast preset doubles Sonar’s score at the same $0.006/query

  • ~7x improvement on agentic tasks: on BrowseComp, where multi-step research is tested, Agent API’s low preset outperforms Sonar Pro by ~7x while maintaining $0.03/query

  • Higher research ceiling: at the top end, Agent API’s high and xhigh surpass Sonar Deep Research, powered by Perplexity’s Search as Code architecture

dsqa-api-announcement
dsqa-api-announcement2165×1412 155 KB

browsecomp-api-announcement
browsecomp-api-announcement2167×1412 165 KB

Every model improvement, automatically

On top of performance, switching to Agent API ensures you will benefit from every model improvement automatically. Our research-backed presets are re-tuned with each frontier and leading open-source model release, leveraging Perplexity’s first-class tools like web_search, finance_search, fetch_url and more.

Read the migration guide

If you’ve already started using with Agent API, keen to hear what you’re building and how it’s comparing to Sonar on cost/latency/quality.

If anything about migrating your Sonar workload isn’t covered in the guide, drop your use case, query volumes, constraints or snippets below and we’ll help map it to the right Agent API preset.

Perplexity API Team