Perplexity Introduces Photon: A Rust-Based Retrieval Engine That Cuts p99 Latency From 800 ms to 65 ms
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/📌 Perplexity自研Rust引擎 p99延遲砍至65毫秒
TL;DR：Perplexity 用 Rust 打造全新檢索引擎 Photon，取代原本的開源 fork，並以 API 形式對外開放 Fast Search 模式。
查詢送出後要等 800 毫秒才看到結果，跟等 65 毫秒的差別，對即時 agent 應用來說幾乎是能不能用的分水嶺。Perplexity 這次直接把底層檢索引擎整套換掉。
🤔 舊引擎撐不住成長中的索引
Perplexity 原本的 AI 原生搜尋堆疊，用的是一套經過自行 fork 的開源檢索引擎。隨著索引規模持續擴大，這套 fork 陸續碰上瓶頸，團隊評估後認為，從零打造一套新引擎，反而比繼續維護 fork 來得簡單也更省成本，於是有了 Photon。
🧩 Broker 分派、Shard 兩階段排序
Photon 的架構是負載平衡器把每個請求導向一個 Photon broker，broker 再把請求分散給一組 shard，並監控逾時狀況。每個 shard 各自完成檢索、初步排序與第二階段排序，broker 最後合併候選結果並抓取關鍵文件欄位。目前完整的網頁索引可以在個位數小時內建置完成。Fast Search 模式則是把 Photon 搭配一套針對 agentic 工作流程調校過的輕量排序邏輯。
📊 同樣任務省下近七成成本，但準確度打了折扣
Perplexity 在 WideSearch、BrowseComp、DSQA、FRAMES、SEAL-0、SEAL-Hard 這 6 個 benchmark、共 3,554 個任務上做了測試：Fast 模式拿下 64.3% 的分數，估計模型加搜尋成本為 59.73 美元；預設模式拿下 64.0%，成本卻要 187.60 美元，Fast 模式便宜了約 68%。
不過代價也反映在內部長尾 benchmark 上：relevance（DCG）從 2.45 降到 2.21，答案可用率從 0.596 降到 0.567，下滑 2.9 個百分點。Photon 本身的單次呼叫延遲為 p50 160 毫秒、p95 230 毫秒。
💡 兩種模式各有適用場景
Perplexity 建議把 Fast 模式用在日常的 agent 迴圈中，而困難、語意模糊的查詢則建議維持使用預設模式。要留意的是，這些延遲數字都是 Perplexity 自行提供、在不同設定下測得，並非同一基準下的公平比較。
🎯 怎麼用：API 開個參數就能切換
Photon 目前不開源，也無法自行架設，只能透過 Perplexity 的託管 API 使用。在 POST /search 呼叫中設定 search_type: “fast”，費用為每 1,000 次請求 1 美元；若使用 Python SDK 0.43.4 或 0.43.5，依官方文件傳入 extra_body={“search_type”: “fast”} 即可。
🔗 來源
- 標題：Perplexity Introduces Photon: A Rust-Based Retrieval Engine That Cuts p99 Latency From 800 ms to 65 ms
- 作者／機構：Asif Razzaq, MarkTechPost
- 連結：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/
#Perplexity #Rust #SearchEngine #Latency #AgenticAI #InfoRetrieval #APIPricing #Benchmarking #SearchAPI #SystemsEngineering
原始資料 MarkTechPost · 收集於 2026-09-30
摘要原文
Perplexity has released Photon , an in-house retrieval and ranking engine written in Rust. It replaces an open-source engine Perplexity had forked for its AI-native search stack. Photon now handles retrieval and ranking for all production traffic. It also powers a new Fast Search mode in the Perplexity Search API. Perplexity reports single-call latency of 160 ms at p50 and 230 ms at p95. Is it deployable? Yes, as a hosted API. Set search_type: "fast" on POST /search and pay $1 per 1,000 requests. Photon itself is not open source, so the engine cannot be self-hosted. The old engine hit 3 limits as the index grew: Perplexity team concluded that building from scratch was simpler and cheaper than maintaining its fork. A load balancer routes each request to a Photon broker. The broker fans out to a shard group and watches for timeouts. Each shard runs retrieval, initial ranking, and second-stage ranking. The broker then merges candidates and fetches key document fields. A full web index now builds in a single-digit number of hours. Fast Search pairs Photon with lighter ranking tuned for agentic workflows. Perplexity tested it on 6 benchmarks: WideSearch, BrowseComp , DSQA, FRAMES, SEAL-0, and SEAL-Hard. Across 3,554 tasks, Fast scored 64.3% at $59.73 in estimated model-plus-search cost. The default preset scored 64.0% at $187.60, so Fast was about 68% cheaper. The trade-off shows up in broader search quality. On internal long-tail benchmarks, relevance (DCG) fell from 2.45 to 2.21. Answer availability dropped from 0.596 to 0.567, a loss of 2.9 percentage points. Perplexity recommends Fast for day-to-day agent loops and the default for hard, ambiguous queries. On Python SDK 0.43.4 and 0.43.5, pass extra_body={"search_type": "fast"} per the docs . All latency figures are vendor-reported under different setups, so they are not like-for-like. Check out the technical details and Fast Search docs . All credit goes to the researcher of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter . Wait! are you on telegram? now you can join us on telegram as well. Need to partner with us for promoting your GitHub Repo OR Hugging Face Page OR Product Release OR Webinar etc.? Connect with us The post Perplexity Introduces Photon: A Rust-Based Retrieval Engine That Cuts p99 Latency From 800 ms to 65 ms appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成