DigestAI news desk

Cut through the AI noise.

Enterprise & Industry4 min read

Perplexity introduces Photon, a Rust-based retrieval engine

Perplexity has launched Photon, a new in‑house retrieval and ranking engine written in Rust that now handles all production traffic and powers a Fast Search mode in its API. Photon reports single‑call latency of 160 ms at p50 and 230 ms at p95, and the engine can be accessed as a hosted API for $1 per 1,000 requests.

1 source

Key points

  • Photon reduces p99 latency from ~800 ms to ~65 ms and runs on 20 % fewer machines
  • Fast Search costs $59.73 for 3,554 tasks, 68 % cheaper than the default preset
  • Photon is a hosted API; pay $1 per 1,000 requests and it is not open source

The new engine replaces an older forked open‑source engine that suffered from high tail latency (p99 near 800 ms), merge spikes, and slow recovery. Fast Search, tuned for agentic workflows, achieved 64.3 % on 3,554 benchmark tasks at an estimated cost of $59.73, about 68 % cheaper than the default preset.

Perplexity notes that Fast Search trades a small drop in relevance (DCG from 2.45 to 2.21) and answer availability (0.596 to 0.567) for lower cost and faster performance, recommending it for day‑to‑day agent loops while keeping the default preset for hard queries.

Full story from MarkTechPost · by Asif RazzaqOpen source ↗

Perplexity Introduces Photon: A Rust-Based Retrieval Engine That Cuts p99 Latency From 800 ms to 65 ms

MarkTechPost · 30 September 2026

Loading the full article…

This text was published by MarkTechPost and written by Asif Razzaq. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗

Topics · follow one to build your own front page

The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.

Comments

via GitHub Discussions

More in Enterprise & Industry

All →

Related stories