Imagine waiting almost a full second every time you hit search. Now imagine that wait shrinking down to a blink. Perplexity just made that real.
The company shipped a brand new retrieval and ranking engine called Photon, built entirely from scratch in Rust, replacing an open-source engine they had been running for years.
The headline number says it all: p99 latency dropped from 800 milliseconds down to just 65 milliseconds.
Think of it like a commute that used to take you 12 minutes and now takes you barely one, with the same route and same effort.
Photon isn't a side experiment either. It now handles retrieval and ranking for all of Perplexity's production traffic, and it powers a brand new Fast Search mode inside their Search API.
The reason behind the leap is simple. Rust is known for being blazing fast and memory-safe, so instead of leaning on a generic open-source engine built for everyone, Perplexity built something custom, tuned precisely to their own workload.
The real win here isn't just a benchmark number. Every query you run now comes back noticeably faster, which matters a lot for AI products that live and die by real-time search.
In the AI race, the line between winning and losing is no longer measured in seconds. It's measured in milliseconds, and Perplexity just moved the goalpost.
🔗 Original Source & Reference: https://www.marktechpost.com/2026/09/30/perplexity-introduces-photon-a-rust-based-retrieval-engine-that-cuts-p99-latency-from-800-ms-to-65-ms/
Published automatically via FeedMind AI Content Pipeline.
Top comments (0)