Perplexity

There is only one thing to weigh: do you need an answer with sources or a conversation partner. Perplexity is about the first.

What people come to Perplexity to do and whether they get it.

Who it is for. For those who look for facts and want to see where they came from: every answer carries links to sites, and a deep report pulls together dozens of sources in a single request. The free tier gives you 3 smart searches a day and one deep report a month (vendor help center, Sep 2, 2026).

Who else. For people who need different models in one window: with Pro for $20 a month, you switch between its Sonar 2 and five outside models, from OpenAI to Google; Max adds Claude Opus 5 and GPT-5.6 Sol. The vendor's full list changes every few weeks and is collected in the questions below.

Who it is not for. For anyone waiting on the voice assistant: across a dozen reviews from May-July 2026 voice mode does not work, and the vendor says nothing about it.

Good fit: section 3 will get you to your first answer. If you're choosing between it and ChatGPT, see the comparison in section 7.

What vibecoding isStarting from scratch: how to set tasks for agents with no coding experience

Two modes on the first screen: Search answers a question, Computer does work. The free offering is in the first.

Search with an answer. You write a question, the service reads websites and returns coherent text with numbered links; each source has its own tile above the answer. The vendor does not count regular searches; smart searches, where the model reasons through several steps, are limited to 3 a day for free. On the free plan you do not choose the model, the service selects it.

Deep report. Since Feb 14, 2025, Deep Research breaks a question into subqueries, gathers dozens of pages, and writes a report with charts. One such report is free per month; Pro and Max have a monthly limit, but the vendor does not publish the number (help, Sep 2, 2026).

Models. On Pro, you switch between its Sonar 2 and five outside models, from OpenAI to Google; Max adds Claude Opus 5 and GPT-5.6 Sol (vendor listings, Sep 6, 2026). Model Council on Max sends one question to several models at once and combines their answers. The list changes, and the vendor itself says the truth is in your account's switcher, so the full list is in the questions below.

Computer. An agent given an entire task: it searches, writes documents and code, creates presentations, browses websites, and connects to Gmail, Slack, GitHub, or Notion; the vendor's connector catalog has more than four hundred. It launched in late February 2026 for Max only, then opened to all Pro users on 12.03.2026. It runs on credits: 100 credits cost $1, and Pro has no monthly credits, only a one-time bonus.

Comet. A browser from the same vendor, built on Chromium with Assistant in a sidebar, reads your tabs and carries out tasks on websites. The free tier and subscription limits are covered in the Comet browser card.

What the vendor itself shows. The official-channel video from Aug 17, 2026, “Search vs Computer,” shows a window with a two-mode switcher; the Apr 10, 2026 video shows Computer on a Mac: the agent's activity feed with terminal commands and a task checklist. The vendor has no complete answer window with sources in its 2026 videos, so the still above comes from its 2025 Deep Research video. We have not run Perplexity ourselves for the task.

You will get your first answer with no setup: the site opens directly, and no sign-in is needed for the first question.

Step 1. Open perplexity.ai and ask a question.

Step 2. Install the app from the App Store or Google Play: there it is published by the vendor itself, version 26.34.0 of 03.09.2026. Samsung installs the app by default, users report in reviews.

Computer. The vendor hands out the Mac and Windows apps from its site, they are not in the Mac App Store. The Comet browser installs separately from perplexity.ai/comet.

What does not work. The voice assistant: "it works with text only", users write in a dozen reviews from May-July 2026, and the earliest App Store complaint is dated 03.03.2025. The vendor has never named either a reason or a date.

Steps verified Sep 6, 2026 against the perplexity.ai help center, the App Store and Google Play listings and user reviews

Among AI chats, Perplexity is third and has lost more than half its share over a year; among search engines, it is not visible at all.

Share. StatCounter for August 2026 puts ChatGPT at 79.4 %, Gemini at 10.9 %, and Perplexity at 4.34 %. It had 11.1 % in October 2025, and in winter 2026 it was overtaken by Gemini. The counter measures referrals from AI chats to other websites, not users. Perplexity does not appear in the same counter's regular search-engine table, where Google has 91.1 %.

Company funding. Annual revenue passed $750 million by August 2026, up from less than $250 million at the start of the year; The Information, via Reuters, names Computer as the driver. The last confirmed valuation was $20 billion in September 2025; in August 2026, Nvidia discussed investment at a valuation above $30 billion. The company head set 2028 as the IPO date (CNBC, Jun 9, 2026).

Audience. The vendor does not publish user numbers. The head of the company cited 780 million queries for May 2025, and Google Play shows more than 100 million installs all-time. The App Store gives it 4.82 across 502 thousand ratings in the US (Sep 6, 2026).

Search demand asks about blocking, while a test says the site opens tool ranking

The Pro price has not changed since June 2024, but the vendor has two pricing pages with different figures, and neither has a «monthly or annual» toggle.

Free. Regular search without a counter, 3 smart searches a day, one deep report a month, basic file uploads. You cannot choose the model, and there are no images or priority support. In June 2024, there were 5 free smart searches a day.

Pro. $20 a month or $200 a year on the pricing page. The perplexity.ai/pro storefront shows "$17 a month", which is the annual price broken down and flagged as a promo. It gives you model choice, deep reports on a monthly limit, images and up to 50 files per project. The vendor does not publish limit numbers for Pro: in June 2024 it promised 600 smart searches a day, in February 2025 "300 and more", now it writes "weekly limit". In May 2026 users complained about 10-20 queries a day, and the vendor replied that the limit depends on how the subscription was obtained.

Max. $200 per month or $2000 per year, with annual billing available only on the site. Beyond Pro, it includes Claude Opus 5 and GPT-5.6 Sol, Model Council, Brain memory, and 10,000 Computer credits per month.

Credits. The Computer agent uses a separate currency: 100 credits cost $1. According to the vendor's documentation, a light task costs 100–350 credits and a heavy one up to 2,275, or from a dollar to more than twenty. Pro has no monthly credits, only a one-time 4,000-credit bonus for 30 days; afterward, packages range from $5 for 500 to $100 for 10,000. Regular search does not spend credits.

API separate. A subscription does not include API access. Its Sonar model costs $1 per million input and output tokens (a token is a piece of a word that models use to count text), Sonar Pro costs $3 and $15, and search as a service costs $5 per thousand requests. The vendor retires the old Sonar API on Sep 27, 2026; its successor is called Agent API.

Intermediaries and promotions. Intermediary markups are not disclosed publicly. Available promotions: one year of Pro for Galaxy owners in the US and a $10 student plan.  .

The product sells trust in sources, and exactly nine organizations are suing it over sources, from newspapers to Reddit.

Data. The policy dated Jul 8, 2026 allows the collection of queries, voice, and browsing history in the Comet browser, and their use «to improve or create products, including our AI models». Training on your data is enabled by default and can be turned off in account settings. The vendor says it does not sell data or send queries to advertisers. The service operates in the US and describes retention as «as long as necessary».

Your content. Ownership remains yours, but the terms give the vendor a free, transferable license to store, modify, and publish uploaded content while it is hosted there. Under the terms, subscriptions are nonrefundable, while the documentation allows 24 hours for a monthly refund and up to three days for an annual one.

Court cases. Dow Jones and New York Post have been suing since Oct 21, 2024; in August 2025, a New York court denied an attempt to move the case. The New York Times filed suit in December 2025, and CNN did so in May 2026 over 17 thousand pieces; the vendor replied: «facts cannot be copyrighted». Britannica, Reddit, and Japanese publishers are also involved. By September 2026, no case had reached a ruling on the merits.

Amazon case. The only case with an appellate ruling: in March 2026, a court barred the Comet agent from buying on Amazon, and on Aug 4, 2026, the appellate court lifted the preliminary injunction and remanded the case to the trial court. The panel's reasoning: the user accesses the site, while the agent only helps them.

Page scrapers. On Aug 4, 2025, Cloudflare accused Perplexity of having its scraper bots change names and addresses to evade website blocks, and removed the company from its verified bot list. The vendor called the publication a marketing stunt. At the same time, it blocked its own site from third-party bots with the same Cloudflare protection: a bare request to perplexity.ai is denied, while pages for this card were captured in a live browser.

Comet. In 2025, three security teams showed how the browser assistant executes commands hidden on a page: Brave on Aug 20, 2025, Guardio in the same month got the agent to enter a saved card on a fake store, and LayerX on Oct 4, 2025 described commands in link parameters. The vendor patched the issues, but a repeat Brave test showed the first flaw was not fully fixed. On Feb 18, 2026, the company decided not to launch ads in answers at all.

The divide is not in the models' intelligence, but in what you want: an answer with verifiable links or a companion for anything.

If you need slides after your search, see how Felo presentationswork: the card covers the path from sources to slides, plans and credit usage.

For a personal chat with search and file analysis, there is Qwen Chat. Its card covers the free tier and usage limits.

How the vendor itself divides them. On its Pro storefront, it says: Perplexity is built around searching the live web with links in every answer, while ChatGPT is a chatbot for conversation; use the former for facts and source checking, the latter for creativity. This is the party's position, not our measurement.

The same engines. An OpenAI model is in both: Perplexity has GPT-5.6 Terra alongside Claude and Gemini. OpenAI also has GPT-6 Astra: on Plus, it is available in Work and Codex, while GPT-6 Pro in regular Chat is for Pro, Business, and Enterprise (OpenAI help, Sep 8, 2026).

ChatGPT paid plans start at $8 per month, Perplexity at $20.

Who else is nearby. Comet from the same vendor takes the assistant into the browser, while Kagi sells ad-free search by subscription.

If you need answers from work email and documents, compare the Dia work browser.

If you need a file for a meeting. Genspark presentations can be built from source documents and edited in a separate editor. Check your plan before you start: exporting slides to a file is on paid plans.

If you need science. Semantic Scholar research search helps you find papers, authors and connections between works. For a table of findings from studies, consider Elicit, and for a summary answer across research papers: Consensus.

In under four years, the answer engine grew into an agent that completes work end to end, accumulated nine lawsuits, and is valued at $20 billion.