Thursday, Sep 24, 2026
1 Perplexity's New Photon Engine Cuts Search API Task Costs by 68% π€ AI Sep 24, 2:07 PM EDT 13/4
The company has launched a Rust-based retrieval and ranking service called Fast Search that returns 95% of search results in 230 milliseconds or less. Driven by a new engine named Photon, the API provides a median latency of 160 milliseconds and reduced cost per task by 68% across six agentic benchmarks while maintaining comparable quality. Internal tests did show a slight decline in long-tail query performance, scoring 0.24 points lower on relevance and 3 percentage points lower on answer availability.
Photon replaces a previous open-source engine and is optimized for agentic workloads at web scale, utilizing approximately 20% fewer machines and storing 2.5 times more data per document. Perplexity built the engine with a small engineering team and $300k in tokens. The Fast Search API is currently the default search for Nous Portal subscribers using the Hermes Agent.