The company’s latest search tool for AI agents returns 95% of results in 230 milliseconds or less, Perplexity announced. The new Fast Search API achieves a median latency of 160 ms and reduced the cost per task by 68% across six agentic benchmarks compared to the company's default preset while maintaining comparable answer quality.

The API is powered by Photon, a Rust-based retrieval and ranking engine that replaces Perplexity's previous open-source system. Built by a small team using $300k in tokens and auto-research loops, Photon dropped internal p99 response times from 800 ms to 65 ms and uses 20% fewer serving machines to store 2.5x as much data per document. Nous Research has already integrated the tool as the default search for Hermes Agent subscribers on the Nous Portal.