Perplexity launches photon: a high-speed retrieval engine for enhanced search performance

Perplexity has unveiled Photon, a proprietary retrieval and ranking engine developed in Rust, designed to replace its previously forked open-source engine. Photon now manages all production traffic and supports a new Fast Search mode within the Perplexity Search API. The system achieves single-call latencies of 160 ms at p50 and 230 ms at p9, with p99 latency reduced from 800 ms to 65 ms. Photon is available as a hosted API, with a $1 per 1,000 requests pricing model, though it is not open source and cannot be self-hosted. The new engine addresses scalability challenges faced by the previous system, which struggled with growing index size. Perplexity opted to build from scratch rather than maintain its fork. Fast Search, which pairs Photon with lightweight ranking tailored for agentic workflows, was tested on six benchmarks, achieving 64.3% performance at a cost of $59.73, compared to the default preset’s 64.0% at $187.60. While Fast Search offers significant cost savings, it slightly impacts broader search quality, with relevance and answer availability metrics declining slightly on internal benchmarks. Perplexity recommends Fast for routine agent tasks and the default for complex queries.