Perplexity 移动 5

• 1. overview (823 chars)

  Perplexity is an AI search product positioned as an "answer engine". Its core selling point is that each answer comes with a clickable source reference, allowing users to verify facts on the spot. This article focuses on its mobile version, which is the direction of the "Perplexity Mobile 5" version marked in daily express: integrating real-time search, voice mode, in-depth research, Spaces space, Comet browser, finance and health modules into a set of "portable research workflow" on the mobile phone. As of mid-2026, the monthly activity of all Perplexity products has exceeded 100 million, and its independent search products handle approximately 780 million queries per month. The valuation will reach US$20 billion in September 2025. The growth is real and fierce. Its moat comes from the citation system, but its shortcomings are weak coverage of Chinese sources, mediocre creative writing, and being locked in a copyright tug of war with a number of news publishers.

• 2. Product background (833 chars)

  Perplexity was founded in 2022 and is headquartered in San Francisco, USA. The founding team mostly comes from OpenAI, Meta and Byte. The company has been betting on the path of "AI search replacing blue links" from the beginning, rather than building a general chatbot. It was valued at US$520 million in January 2025 and has risen to US$20 billion by September 2025, a 38-fold increase in about 20 months. In 2026, it further shifted its business model from advertising to pure subscription. In February, it directly cut off the entire advertising business and canceled sponsored answers, which resulted in a significant increase in product credibility. The mobile terminal is Perplexity's fastest-growing entry point - the cumulative downloads of iOS applications exceed 80 million, and the App Store rating is about 4.7 stars. Many reviewers believe that "it is an AI interface designed from scratch for mobile phones, rather than shoehorning the desktop version into the small screen."

• 3. Product functions and usage experience (2843 chars)

  The feature density of this generation of Mobile 5 is significantly higher than that of earlier versions. Real-time search and citation are the base: you ask questions in natural language, and it crawls the current web page in the background and synthesizes it to give a direct answer with numbered citations. There are no ads or SEO intrusion in the whole process. Voice Mode has been launched on iOS and Android at the same time, supporting playback at up to 2x speed, suitable for commuting or multi-tasking scenarios, and listening to answers while walking. Deep Research can run a multi-stage research process directly on your mobile phone, perform dozens of searches, read source documents, identify gaps and complete them, and finally produce a long structured report with dense citations. Results for complex questions can be produced in about 1 to 3 minutes. It is available to Pro and Max users, and free users have a small daily quota.   Spaces turns Perplexity from a one-time search tool into a sustainable research system: create a space on your mobile phone, set exclusive AI commands, upload PDFs or documents and they can be referenced in any query for a long time, all sessions are synchronized in real time across devices, and Pro/Max can also invite collaborators to build together. Focus mode (Focus) allows you to limit the search scope to academic, YouTube, Reddit, news or Wolfram Alpha, which is much more accurate than general search when doing serious research. In early 2026, iOS-specific quiz and flashcard functions will also be launched. By typing "Help me make a set of geography questions and answers", interactive learning materials can be generated, which can be used by student groups.   Comet is Perplexity’s own AI browser. It was first launched on Mac in July 2025, and on iPhone and iPad in March 2026. After that, it launched a wave of 8 iOS updates (one-touch dialing/FaceTime, iPad sidebar redesign, financial in-depth reading opened in real tabs, pictures dragged into the assistant, etc.). The soul of Comet is "page awareness": it knows which tab you are looking at, and can ask questions, summarize, and compare prices on the current web page content at any time without copying and pasting. It should be noted that Perplexity will collect browsing and search records in Comet for ad targeting, and there is currently no one-click exit switch in the app. Privacy-conscious users are more suitable to use the main app instead of using it as their default browser.Several major updates in 2026 will push the mobile terminal further into the "production platform": in June, memory across personal accounts was added (available to everyone, and can be turned off in the settings), inline financial charts (search "AAPL" to directly access stock price charts, ETF positions and analyst ratings), Health health connector (can access millions of medical institution records and Fitbit/Google Fit, combined with Computer intelligence to generate personalized health plans); April Deep Research and Pro Search It can directly produce presentations, tables, dashboards and websites. At the model level, Max users can use the Model Council to run the same problem through multiple cutting-edge models such as GPT, Claude, and Kimi at one time and compare the similarities and differences side by side. This capability is still unique among consumer products.

• 4. Pricing and business model (1113 chars)

  Perplexity uses a clear free + subscription tier. Free version: unlimited basic searches, citations for every question, about 3 Pro searches and 3 in-depth studies per day, automatic selection of core models, and focus mode available, which is enough for daily verification. The Pro version is $20/month (annual payment is about $17/month): 200 Pro queries per week, 20 in-depth studies per month, file uploads, multi-model switching, voice mode, Spaces, Labs asset generation, and $5 monthly API quota. The price/performance ratio is the highest for most researchers. Max version $200/month: Based on Pro, it has additional Perplexity Computer autonomous agents, higher in-depth research quota, priority computing power and early access to the latest models. On the enterprise side, there are Enterprise Pro (about $40/seat/month) and Enterprise Max (about $325/seat/month), which bring compliance commitments such as SSO, audit logs, and data not being used for training. A common complaint is that in early 2026, the in-depth research quota of Pro will be cut from hundreds of times a day in the early days to 20 times a month. Heavy users are forced to go to Max, and the experience gap is obvious. Education users who have been certified can get Education Pro for free (some channels show a discount of US$10/month).

• 5. User reviews (1584 chars)

  Positive voices are concentrated at three points. First, the mobile interface is considered to be one of the best-designed AI apps currently. It is smooth to operate with one hand, responds quickly, and questions naturally. The high score of 4.7 stars on iOS reflects true satisfaction. Second, the citation system is the clearest competitive advantage over pure LLM - checking what happened last week can provide credible answers to the source, and each item can be clicked back to the original text for verification. The risk of hallucinations is significantly lower than ChatGPT or Claude. Third, Spaces and Model Council do not have standard search counterparts, which is especially useful when doing competitive product analysis, market research, and continuous due diligence. Chinese reviews also recognize its global dominance of multi-language information sources. By taking a photo of a restaurant or a used car, you can compare prices and check for common problems, which is much faster than browsing review websites.   The negative feedback was equally clear. Source quality varies: Although the citation is good, the answer will be distorted when the underlying web page itself has low-quality or SEO content. The Chinese context understanding is inferior to local products (such as Secret Tower Search), and it is not as accurate as domestic tools when encountering slang or slang in very small Chinese communities. Creative writing and emotional expression are average, and novel or story generation is not as delicate as a dedicated model. ChatGPT/Claude is stronger for users who rely on complex code. The early Comet iOS version also had obvious bugs such as "deleted conversations will be resurrected" and "recovery after clearing the record and restarting". Although they were fixed in subsequent updates, they exposed first-release quality control issues.

• 6. Industry evaluation (912 chars)

  The industry generally regards Perplexity as the benchmark for the "research-based AI search" category. The evaluation agency gave it a score of 4.7 to 4.8. The core judgment is that it has found a balance between accuracy, timeliness and traceability, and its status as a universal "answer engine" is difficult to shake. The consensus compared with competing products is: for Google, Perplexity gives direct answers + references instead of link lists; for ChatGPT/Claude, it searches web pages every time and shows evidence, while the latter's answers come from training data and need to be verified separately; for Gemini, its reference structure is more complete; for domestic secret towers and DeepSeek, it dominates the breadth of global information sources, but lags behind in Chinese localization and code reasoning. In February, Perplexity introduced the DRACO accuracy benchmark (approximately 79.5%) jointly developed with Harvard to endorse the quality of its answers, which also reflects the rising industry demand for "quantifiable credibility of AI answers."

• 7. Controversies and Risks (1248 chars)

  Copyright litigation is Perplexity's biggest external risk right now. The New York Times sued at the end of 2025. In May 2026, CNN filed a 54-page complaint in the Federal Court of the Southern District of New York, accusing it of grabbing and redistributing more than 17,000 CNN reports, pictures and videos. Some of the answers were "word for word" similar to the original text. It also pointed out that its Comet Plus had hinted that it could access CNN paid content (the two parties negotiated authorization in 2025 but failed). Perplexity’s response has always been “You can’t copyright facts.” In addition to CNN, News Corp/Dow Jones, New York Post, Chicago Tribune, Encyclopedia Britannica, Merriam-Webster Dictionary, Reddit, Yomiuri Shimbun, Amazon (for AI agent login shopping), etc. have all filed lawsuits against it. Most of the cases are still in the early stages, and the core "whether AI crawling constitutes infringement" has not yet been decided. On the other side, Times, Gannett, Le Monde, Der Spiegel, etc. have signed licensing agreements with Perplexity, and the industry is bifurcating along the two paths of "litigation red lines vs. licensing exchange volume". For ordinary users, these lawsuits will not affect use in the short term, but may change which publishing sources can appear in the answers in the long term. In terms of privacy, Comet collects browsing history for ad targeting and lacks an exit switch, which is another problem that needs attention.

• 8. Applicable groups and usage suggestions (917 chars)

  People who are most suitable for using Perplexity mobile app: students and researchers (check facts before writing a paper and bring references), reporters and analysts (quickly integrate real-time reports from multiple sources), professionals to check industry and company intelligence, and people who want to get rid of advertisements and do homework on their mobile phones. The free version is equivalent to the "upgraded version of Google" for most Chinese users, so there is no need to rush to pay; only when you search more than about 5 times a day, or really need to do in-depth research and collaboration with Spaces, consider the $20/month Pro. Not suitable for: those who want daily chatting in Chinese (ChatGPT is recommended), those who want to write creative literature, those who need assistance with complex codes, and those who rely heavily on Chinese Internet slang for social media operations. As for alternatives, if you want in-depth Chinese reading, choose Secret Tower; if you want code and logic, choose DeepSeek; if you want an all-around assistant, choose ChatGPT.

• 9. Summarize (637 chars)

  The Perplexity Mobile 5 generation pushes the "portable answer engine" into the "portable research workflow": voice, in-depth research, Spaces, Comet, finance and health modules are strung together in the same mobile phone to form a closed loop. The citation system is still its most difficult moat to be copied. The price is that the Chinese source is still weak, the Pro quota has been cut and forced to increase to Max, and there are two dark clouds of copyright and privacy hanging over the head. For the vast majority of people who want to obtain global information efficiently, it is still the most worth installing AI App at present - just don’t regard it as the only source of truth. Remember to click back to the original text to verify important conclusions.

No in-depth report yet

• 

  LenniHarju

  —

  Model Council runs the same question at one time and compares three cutting-edge models side by side. When doing strategic analysis and investment research, you can directly see the differences between the models. This is really not available in other consumer products. In the past, I had to open three apps for cross-validation, but now I can read them all in one interface. The time saved was enough for me to run two more sets of backtests.

---

• 

  Katherine.Wright_2022

  —

  I use Perplexity on my phone more than ChatGPT. Its logic is not to give you an answer, but to search more than a dozen sources and then summarize it. Each conclusion has a clickable source, and you can verify it on the spot by clicking on it. This kind of credibility cannot be given by pure chatbots relying on training data. The scenario of checking the information is completely dimensionality reduction.

---

• 

  fkk1nr6dh

  —

  Voice Mode is great for listening to answers while walking during commuting. It can also adjust the speed twice, which is much more comfortable than staring at a small screen. Moreover, it supports mixed questions in Chinese and English. If you just ask a question while walking or cooking, you can hear a spoken summary with quotes. It completely liberated me from the small screen of my mobile phone. I strongly recommend that drivers try it.

---

• 

  LiquidityKingWood

  —

  Searching for things is ten times easier than Baidu. If you take a photo of a roadside restaurant, you can find the recommended dishes per person in the reviews, which is much faster than browsing review websites.

---

• 

  MJimenez0

  —

  Spaces turns scattered search into a sustainable research space. You can get cited every time after uploading a PDF. It is indispensable when writing competitive product analysis and market research. Team collaboration can also share context.

---

• 

  Sarah_Walker_7514

  —

  The iOS interface is truly designed from the ground up for mobile phones. It is smooth to operate with one hand, responds quickly, and questions naturally. The high rating of 4.7 stars is not just a brush. Compared with those AI apps that force the desktop version into a small screen, its mobile experience is obviously more convenient. I can complete most queries with one hand during commuting. This is the fundamental reason why I am willing to open it every day.

---

• 

  梅花_19

  —

  The Comet browser is convenient for being able to ask questions about the current web page at any time, but it collects browsing history for advertising targeting and does not have an exit switch. People with strong privacy awareness are more at ease using the main app as a search engine.

---

• 

  阎丹刚

  —

  The coverage of Chinese sources is still weak. It is not as accurate as Secret Tower when searching for slang in domestic niche communities and very niche Chinese content. Users who rely heavily on the Chinese Internet must use other tools. Don’t use it as the only information entrance, otherwise it will be easy to be biased by English sources.

---

• 

  MiningMoleStewart

  —

  A lot of people on Trustpilot criticized automatic renewals and hidden deductions. Deep Research cut its quota from 600 times a day to 20 times a month without notifying users in advance. This silent plan change is really unpopular. Be sure to read every change email before subscribing, and annual paying users should pay more attention to the refund terms. Don’t wait until you are repeatedly deducted to find out that there is no way to complain.

---

• 

  AJenkinsSr

  —

  The shortcomings are obvious: long conversations occasionally mix the content of different threads. I also lost a large section of completed analysis. Heavy users have encountered this several times when working on long projects. Important conclusion: Remember to save another copy to be safe. Don’t bet all on its session history, otherwise you won’t be able to cry when you clear it one day.

---

• 

  王华

  —

  The free version is sufficient for most people: unlimited basic searches, citations for every question, no ads, and is completely sufficient for daily verification. Don’t rush for the tuition fee. It’s only worth buying Pro if you search more than five times a day or frequently write reports with references. Otherwise, the free version is just an upgraded version of Google, and there’s no need to spend twenty dollars a month.

---

• 

  ANjen_s

  —

  The quality of the source is still uneven. When the underlying web page itself is SEO hydrological, the answer will also follow. You cannot believe it all. Before making key decisions, be sure to read the original text before drawing a conclusion.

---

• 

  SLewisSr

  —

  I have autism, and I once uploaded a screenshot of a conversation that I couldn't understand at all. It actually helped me correctly understand the social subtext and overtones, and it also pointed out that the other party was actually politely rejecting the conversation. No other app has done this kind of thing. It is very helpful for people with social dyslexia like me. I have already recommended it to several friends in the same circle.

---

• 

  孔彤_1

  —

  The mobile version is generally good, but you may occasionally experience hallucinations when asking about some particularly unpopular technical parameters. Remember to click back to check the original link for important conclusions.

---

• 

  Abigail_Gonzales520

  —

  Perplexity has completely replaced Google as my main source of daily information search. Every answer comes with a clickable source. This citation mechanism is really a dimensionality reduction blow when doing literature reviews and academic verification. You no longer have to go through dozens of original texts to check the sources one by one. The efficiency of fact-checking before writing a paper has at least tripled.

---

• 

  8zawqa

  —

  The customer service is basically non-existent. The annual payment was deducted repeatedly, and no one responded to several emails for two weeks. This kind of billing experience is really discouraging; you must read the automatic renewal terms clearly before subscribing, don’t be like me who was silently charged only to find out that the process of canceling the refund was also ridiculously long.

---

• 

  Cynthia_Lewis5204

  —

  Some summaries are still superficial. Fortunately, the query function can help me dig down to the actual data. Using more rounds is more reliable than just looking at the first screen.

---

• 

  Betty_Murray_2020

  —

  The best thing is that I can switch models between Claude and GPT at any time. For me to write technical documents, the $20 a month of Pro is really worth it.

---

• 

  蝴蝶617

  —

  Deep Research is really useful for dealing with complex topics. After running, it can produce long reports with dense citations, which is much more efficient than manual search. However, when I am stuck on a deadline, I am often annoyed by the new quota limit. I still have to leave the heavy work until the beginning of the month, and I am tied up at the end of the month. I hope the official will relax the quota a little more.

No reviews yet