Zubnet AI学习Wiki › Perplexity
公司

Perplexity

又名: AI-powered search engine, Sonar API
把实时网页搜索和语言模型推理结合起来,给出直接、有来源的答案而不是一列链接的 AI 搜索引擎。一代人以来对 Google 搜索主导地位最显眼的挑战。

为什么重要

Perplexity 是十多年来对 Google 搜索主导地位最可信的挑战,证明一个 AI 原生的答案引擎能为信息寻找查询提供根本更好的体验。他们把 RAG 范式作为消费产品推广,展示了把实时网页搜索与 LLM 推理结合,产出的结果比两种技术单独使用都更有用、更可信。他们的快速增长迫使 Google、Microsoft 和每个其他搜索玩家重新思考大语言模型时代的搜索引擎该是什么样。

Deep Dive

Perplexity was founded in August 2022 by Aravind Srinivas, Denis Yarats, Johnny Ho, and Andy Konwinski. Srinivas, the CEO, is a former research scientist at OpenAI and DeepMind who did his PhD at UC Berkeley under the legendary Pieter Abbeel. Yarats came from Meta AI Research, Ho from Quora, and Konwinski was a co-creator of Apache Spark and co-founder of Databricks. They saw a gap: search engines returned ten blue links and left you to do the reading, while chatbots hallucinated and gave no sources. Perplexity would combine real-time web search with language model reasoning to give you a direct, cited answer. The company raised early backing from Jeff Bezos, NVIDIA, and Andreessen Horowitz, and by late 2024 had reached a $9 billion valuation after raising over $900 million in total funding.

The Product: Answer Engines Over Search Engines

Perplexity calls itself an "answer engine" rather than a search engine, and the distinction matters. When you ask Perplexity a question, it searches the web in real time, reads the relevant pages, synthesizes the information, and presents a coherent answer with inline citations you can click to verify. The product comes in two tiers: the free version uses their own models and a standard search index, while Perplexity Pro lets you choose between multiple frontier models (Claude, GPT-4, and others) for the reasoning step and includes more complex research capabilities. Their "Focus" modes let you target searches at specific domains — academic papers, YouTube videos, Reddit discussions, code repositories. The Pro Search feature breaks complex questions into sub-queries and synthesizes findings across multiple searches, effectively doing 10 minutes of research in 30 seconds.

The Google Problem

Taking on Google in search is the kind of ambition that gets you either admired or laughed at, and Perplexity has experienced both. Their advantage is that the AI-native format — direct answers with sources — is genuinely better for many queries than a list of ad-laden links. Their disadvantage is that Google has effectively infinite resources, decades of search infrastructure, and launched their own AI Overviews feature that does something similar right on the search results page. Perplexity has also drawn criticism for how its web crawler scrapes content from publishers, with some news organizations accusing the company of reproducing their journalism without adequate attribution or compensation. Forbes, Condé Nast, and others have raised objections, and Perplexity responded by launching a revenue-sharing program for publishers — a smart move, though critics note the terms heavily favor Perplexity.

Business Model and Growth

Perplexity monetizes through Perplexity Pro subscriptions ($20/month or $200/year) and an enterprise tier for businesses. They have also begun experimenting with advertising, introducing sponsored follow-up questions alongside organic results — a move that drew some backlash from users who valued the ad-free experience. The company reported over 15 million monthly active users by mid-2024 and continued growing rapidly through 2025. Their API business lets developers plug Perplexity's search-augmented generation into their own applications, creating a second revenue stream. The economics are challenging — each query involves both a web search and an LLM inference call, making the cost per query significantly higher than traditional search — but the bet is that users will pay for answers that actually save them time.

Where It Gets Interesting

Perplexity occupies a fascinating position in the AI landscape. They are not a foundation model company — they use other companies' models for reasoning — but they are not a thin wrapper either, because their search infrastructure, indexing, and retrieval pipeline are substantial pieces of engineering. They have also started training their own models for specific tasks, blurring the line further. The company represents the strongest evidence yet that the search paradigm established by Google in 1998 is genuinely vulnerable to disruption. Whether Perplexity specifically becomes the successor, or merely the proof of concept that showed everyone else the way, the era of "ten blue links" as the default interface to the internet is clearly ending.

相关概念

In The News

Perplexity ditches Google chase, goes after Mint with banking agents
Apr 10, 2026
Perplexity Caught Sharing Every Chat With Google and Meta
Apr 03, 2026
← 所有术语
← Parameters Perplexity (Metric) →
ESC