Perplexity Review 2026: The Answer Engine That Became an Agent Company -- and Started Charging Credits
Perplexity began as the AI tool that put sources next to every answer. In 2026 it is also Computer, an agent that does multi-step work for you; Comet, an AI browser; and an API platform. It is also, according to its own help center, a product where the agent part runs on credits at 100 credits per dollar, with Pro subscribers getting no monthly allowance. Meanwhile an independent audit this month found a large share of its numeric citations did not support the number they were attached to, and its Trustpilot score is 1.4 out of 5. This review checks what is true today, using Perplexity's own help center and docs, and then adds what real users and outside auditors say.
| Ownership | Independent, venture-backed |
| Free tier | Real -- basic searches, very few Pro Searches |
| Paid plans | Pro $20, Max $200 per month |
| Agent credits | 100 = $1; Pro has no monthly allowance |
| Trustpilot record | 1.4/5, 825 reviews |
Still the best interface for finding sourced answers fast. Verify the sources, and read the credits page before you pay for the agent.
"Perplexity's original idea is still its strongest: ask a question, get a short answer with numbered sources you can open. No other mainstream assistant puts that front and centre. The problem is that the company has moved its growth story to agents and credits, while the thing most people actually pay $20 for, trustworthy cited answers, is where outside testing finds the cracks. A citation is a claim that the source says this. It is only useful if you check it."
My read: Free is a good way to see whether the cited-answer style suits you. Pro at $20 makes sense if you do research daily and will click the sources. Max at $200 makes sense only for people who will actually spend agent credits. And if you mainly want a general assistant or a coding partner, other assistants JAVIS reviews cover more ground for the same money.
Ask, research, build, browse, and delegate -- all under one account.
As of this check Perplexity offers Search (fast cited answers, with a "Best" auto-model option and manual models on paid plans), Deep Research (long reports with charts), Learn Mode, Model Council (several models answering at once; Max only), Create files and apps, Comet (the AI browser), and Computer, an autonomous agent for long-running work that is billed in credits. Computer arrived as a major product in February 2026 and has since gained Brain (a memory system), email delivery (send tasks to computer@perplexity.com), Projects (formerly Spaces), a Windows "Personal Computer", and hybrid Mac-plus-cloud processing. For developers there is the Sonar API, a Search API and an Agent API.
One piece of context: Perplexity is independent and has no parent company. The naming change worth knowing is that "Spaces" became "Projects" on Jul 30, 2026.
A small share of chatbot traffic, a big share of the "answer engine" conversation, and an unusually legal-heavy year.
Third-party: Similarweb May 2026 (138M visits, via press) and FirstPageSage Oct 2026. Traffic, not users.
Third-party trackers; $20B from a Sept 2025 raise. Talks at ~$30B are reported, not confirmed by Perplexity.
Tracker figure, not a Perplexity disclosure. Treat as an estimate.
Copyright and trademark suits remain pending; Amazon's Comet case was decided on appeal Aug 4.
| Date | Event | Source |
|---|---|---|
| 2024-10 | Dow Jones sues over scraped journalism | Press / court filings |
| 2025-12-05 | New York Times sues in SDNY; amended complaint filed Mar 20, 2026 | NYT 10-Q, press |
| 2026-02 | Computer agent launched (usage-based credits layered on subscriptions) | Tracker reports |
| 2026-03-09 | Court grants Amazon injunction against Comet on Amazon account pages | CNBC, Ropes & Gray |
| 2026-05-18 | Pro users report cut limits; Perplexity blames promo-code fraud | Android Authority |
| 2026-08-04 | Ninth Circuit vacates the Amazon injunction (user, not agent, "accesses" the site) | Jones Day, Ropes & Gray |
| 2026-09-02 | Third-party audits of Perplexity citations hit Hacker News | HN |
| 2026-09-15 | Press: Comet "Control browser" now needs Computer credits | PiunikaWeb |
| 2026-09-21 | Computer update: effort modes, GPT-6 Astra for eligible Pro/Max, portable and hybrid Computer | Release notes (via Releasebot) |
The loudest complaints are about credits, limits, and whether the citations hold up.
1.4 out of 5; 8% five-star, 83% one-star. "All activity in computer mode costs credits which are $$. Very sneaky and expensive." Another: "The company does not have any customer support... you are permanently on your own." Minority praise: search and structured answers for research and coding.
Asked 310 factual questions about 210 tech companies and checked 1,826 citations on numeric claims: 34.7% pointed to pages that would not open or lacked the cited figure. Scored per claim, 14.4% failed. Only 1.3% were dead links; 16.1% were gated pages. The authors note one snapshot, English tech questions only, and that "a claim can be true with a bad citation."
On the audit: "I noticed this personally... Checked, and it referenced a Reddit article and various other less reliable sources." On value: "I'm not sure what the value of perplexity is these days," and "With all major llms looking up the web ... I'm not sure there's any use case left for perplexity." These are small samples and strong opinions, not a survey.
Free, Pro, Max and two Enterprise tiers -- plus a separate credit meter for Computer.
Free
Practically unlimited basic searches.
- Very limited Pro Searches
- Auto-selected model only
- No image generation or premium support
Pro
Advanced models, research, uploads.
- Manual model choice, Deep Research
- Create files and apps (limited)
- One-time 4,000-credit bonus, no monthly credits
Max
$2,000/yr on web.
- 10,000 Computer credits per month
- Model Council, Brain, early access
- 35,000-credit one-time bonus
Enterprise Pro / Max
Pro from $40/seat; Max ~$325/seat (tracker).
- Data never used for training
- Admin controls, billing
- Max adds SCIM, audit logs, retention
| Plan | Price (USD) | Computer credits | Source |
|---|---|---|---|
| Free | $0 | None | Help center |
| Education Pro | $10/mo (verified students and educators) | Includes Computer access | Help center |
| Pro | $20/mo (trackers) | No monthly set amount; 4,000 one-time, expires in 30 days | Credits help article; trackers for price |
| Max | $200/mo or $2,000/yr | 10,000/month; 35,000 one-time bonus; default $200/month spend cap (adjustable to $5,000) | Help center |
| Enterprise Pro | From $40/mo or $400/yr per seat | Set monthly amount per plan | Help center |
| Enterprise Max | ~$325/seat (trackers) | Higher monthly amount | Trackers |
Pick by whether you want cited answers or an agent.
Free. Use it on questions where you can check the sources.
Pro at $20 for manual model choice and Deep Research.
Check Max ($200, 10,000 credits a month) or buy credits; estimate the credits first (100 = $1).
Enterprise Pro from $40 per seat; Max tier for audit logs and retention controls.
Education Pro at $10 per month with verification.
Compare ChatGPT, Gemini or Claude before paying.
The product's real identity: an answer with its sources attached.
Search is still the heart of Perplexity. The default Best mode picks a model for each query; paid plans can choose models including Perplexity's own Sonar and third-party models, and Perplexity's help center says one subscription includes "all our latest and most powerful AI models" from OpenAI, Anthropic, Google and others (its list was a September 4 snapshot and it warns the model selector is the source of truth). Deep Research chooses models automatically and produces long reports with more sources and charts. Perplexity's own Pro description says Pro gets "10x as many citations per answer for greater reference depth and transparency." That promise is exactly what the September audit tested, and it is why I recommend treating every number as a lead until you open the link.
| Mode | What it does | Plans |
|---|---|---|
| Search / Best | Fast cited answers; model chosen for you | All, no quota on "Best" |
| Pro Search | Multi-step search; manual model choice | Very limited on Free, extended on Pro |
| Deep Research | Long reports, models picked automatically | Extended on Pro; limits vary by plan |
| Learn Mode | Interactive explanations and quizzes | Named in Education Pro; check your selector |
| Model Council | Several models on one question, synthesized | Max and Enterprise Max, web |
- Showing sources by default is a real design advantage over assistants that hide them.
- Ask searches do not consume credits, per Perplexity's credits article.
- Perplexity published an open-source agent-security suite (Numbat) in July.
- A Sept 2 third-party audit found 34.7% of numeric citations failed to show the cited figure; open the link before you rely on a number.
- Model names in Perplexity's help text lag its release notes, so check the selector.
The official "Perplexity Computer" video, which supports the agent section below.
Published by the official "Perplexity" YouTube channel (youtube.com/@perplexity-ai).
Ask, read the sources, and only then act on the answer.
The human-review step has two parts in Perplexity. One is the citation check. The other is spend: Perplexity's credits page says each Computer thread has a Usage dropdown showing credits used, and your account's usage page shows bonus, plan and purchased balances. Looking at that after your first few tasks is the cheapest way to learn what a task really costs.
Agents are where Perplexity is spending its effort, and where the billing got complicated.
Computer is Perplexity's agent for long-running work: research, code, documents, connected tools. Perplexity's credits article says credits are consumed only by Computer, that paused tasks resume when credits return, and that auto-refill triggers at 500 credits for Pro and 2,500 for Max. Recent additions include Brain (a self-improving memory system, free during research preview on Max), Computer by email, Personal Computer on Windows, hybrid Mac-plus-cloud processing for sensitive work, effort modes (Light to Ultra), and, per Sept 21 release notes, GPT-6 Astra for eligible Pro and Max users. Comet is the AI browser, free to download, with an assistant that completes browser tasks. On Sept 15, PiunikaWeb reported that Comet's "Control browser" feature, once a Pro selling point, now draws on Computer credits; I found no Perplexity statement or help page confirming that, so treat it as reported.
| Feature | Plan | Cost model |
|---|---|---|
| Computer tasks | Pro, Max, Enterprise | Credits (100 = $1); light tasks ~15-70 credits |
| Brain memory | Max (research preview) | Free during preview |
| Comet browser | All plans (download is free) | Browser control reportedly credit-gated (press) |
| Computer by email | Eligible plans | Same credits |
| Personal Computer (Mac / Windows) | Eligible plans | Same credits |
- Delegating a defined research-and-document job and getting a file back, including by email.
- The Ninth Circuit ruling in August removed an injunction that had blocked Comet from Amazon account pages.
- Pro has no monthly credit allowance, so heavy Computer use means buying credits.
- An agent that acts in your browser and connected apps is a higher-permission surface; I have not hands-on tested Computer or Comet.
Microsoft 365, developer tools and an Agent API with MCP support.
Perplexity's Sept 21 release notes list a Microsoft 365 connector (SharePoint, OneDrive, Excel), builder connectors (GitHub, Vercel, Datadog), new connectors (Omnisend, Higgsfield, Evernote), Slack workflows and an Enterprise analytics API. Its Aug 13 blog post describes the Agent API as one endpoint for web search, URL fetching, code execution, MCP, and finance and people search. If you work with MCP, our Claude MCP and Connectors guide explains the protocol that standard rests on.
| Integration | What it does | Source |
|---|---|---|
| Microsoft 365 connector | Search SharePoint, OneDrive, Excel workbooks | Release notes, Sep 21 |
| Builder connectors | GitHub, Vercel, Datadog for developers | Release notes, Sep 21 |
| Agent API (incl. MCP) | Web search, fetch, code execution, finance and people search | Perplexity blog, Aug 13 |
| Skills marketplace | Install reusable capabilities for Computer | Release notes, Sep 21 |
| Send or cc computer@perplexity.com to run a task | Perplexity blog, Aug 18 |
Subscription, credits and API are three separate meters.
| What you're doing | What you pay |
|---|---|
| Asking searches in the app | Covered by your plan; Ask searches use no credits |
| Running Computer tasks | Credits: 100 = $1; Max includes 10,000 a month; Pro none monthly |
| Sonar API | $1 per 1M input and output tokens plus request fees of $5-$12 per 1,000 |
| Sonar Pro API | $3 input / $15 output per 1M tokens plus $6-$14 per 1,000 requests |
| Sonar Deep Research API | $2 / $8 per 1M tokens, plus citation, reasoning and $5 per 1,000 search charges |
| Search API | $5 per 1,000 requests standard; $1 per 1,000 fast |
| Agent API tools | Web search $0.0025 per call; fetch URL $0.0005 |
Enterprise data is protected by default; consumer data is not.
| Control | What's in place |
|---|---|
| Enterprise Pro data | Perplexity's help center states data is never used for training; Trust Center access |
| Enterprise Max extras | SCIM, audit logs, configurable data retention, insights dashboard |
| Consumer data (Free, Pro, Max) | Secondary sources report queries are used to improve models by default; an "AI data retention" toggle in account Preferences turns that off for future data |
| Agent safety | Open-source Numbat suite detects risky agent behaviour on macOS, Linux, Windows (Perplexity, Jul 29) |
| Legal exposure | Pending copyright and trademark suits (NYT, Dow Jones); Amazon injunction vacated Aug 4 |
Before you paste anything sensitive into a consumer plan, check the AI data retention toggle. I could not read Perplexity's privacy policy directly, so confirm the current wording in the app.
Best for people who research and will verify; weaker for people who want a hands-off general assistant.
| Situation | Why Perplexity fits | Likely plan |
|---|---|---|
| Student or educator | Learn Mode, Education Pro at $10 | Education Pro |
| Analyst or journalist | Cited answers and Deep Research, with manual source checks | Pro |
| Solo founder or builder | Computer for builders, GitHub and Vercel connectors, Agent API | Pro or Max |
| Enterprise team | Admin controls, no training on data, audit logs on Max | Enterprise Pro / Max |
| Developer needing search in an app | Sonar, Search and Agent APIs, pay per use | API |
| Someone who wants chat, voice, files and coding in one | Not the strongest fit; compare ChatGPT and Gemini | Free to test |
Four real gaps, not manufactured ones.
A Sept 2 third-party audit found 34.7% of numeric citations failed to show the cited figure (14.4% by claim). It has stated limits, but it matches what HN commenters describe.
Computer runs on credits at 100 per dollar; Pro has no monthly allowance, and reviewers call this "sneaky and expensive". Press also reports Comet browser control now draws on credits.
In May, Pro users reported tighter limits and Perplexity attributed differences to promo-code fraud enforcement. Trustpilot reviewers say support is automated only.
NYT and Dow Jones suits are pending, and a 519-point HN thread alleged Perplexity cites mass-produced "best software" pages. That thread links a third-party report I did not independently verify.
Perplexity is the search-first specialist. Others cover more ground.
| If you mostly need... | Compare Perplexity with... |
|---|---|
| An all-in-one assistant: chat, voice, files, agents | ChatGPT (JAVIS review) |
| Google apps, storage and a big free tier | Google Gemini (JAVIS review) |
| Careful writing and coding | Claude |
| A coding agent rather than a search engine | OpenAI Codex, GitHub Copilot |
| MCP-based connectors for your own tools | Claude MCP & Connectors guide |
| Consumer tier (US, monthly) | Perplexity | Gemini | ChatGPT |
|---|---|---|---|
| Standard paid | Pro $20 | AI Pro $19.99 | Plus $20 |
| Heavy-use | Max $200 | AI Ultra $99.99 / $199.99 | Pro $100 / $200 / $500 |
| Agent spend model | Credits (100 = $1) | Plan limits plus AI credits | Plan allowance plus credits |
No affiliate relationship shapes this review.
JAVIS has not established a publisher affiliate program with Perplexity. Every CTA in this article points to the official product.
Go to the official page.
Open PerplexityRead the alternative that matches your real question.
Choose the next articlePerplexity sits next to the big assistants and the connector ecosystem.
Card images are official assets from each product's own site or from the already-published JAVIS target page.
Questions people are actually searching right now
How much does Perplexity cost?
Free is $0. Pro is about $20 a month and Max is $200 a month ($2,000 a year on web), per Perplexity's help center and pricing trackers. Enterprise Pro starts at $40 per seat per month. Education Pro is $10 a month for verified students and educators.
Is Perplexity free?
Yes. Free includes practically unlimited basic searches and a very limited number of Pro Searches, with the model chosen for you.
What are Perplexity credits?
Credits power Computer, Perplexity's agent. 100 credits equal $1. Max includes 10,000 a month, Pro has no monthly amount, and new subscribers may get a one-time bonus that expires in 30 days. Ask searches do not use credits.
Are Perplexity's citations accurate?
Not always. A Sept 2, 2026 third-party audit of 1,826 numeric citations found 34.7% pointed to pages that did not open or lacked the figure. The authors list limits, but you should open the source before relying on a number.
Does Perplexity train on my data?
On Enterprise Pro, Perplexity says data is never used for training. On consumer plans, secondary sources say queries are used by default unless you turn off AI data retention in Preferences.
What is Comet?
Perplexity's AI browser, free to download, with an assistant that completes browser tasks. Press reported Sept 15 that browser control now uses Computer credits; I could not confirm this on a Perplexity page.
Was Perplexity blocked from Amazon?
A court granted Amazon an injunction against Comet on Mar 9, 2026, but the Ninth Circuit vacated it on Aug 4, 2026, holding on the current record that the user, not the agent, accesses the site.
Is Perplexity better than ChatGPT or Gemini?
It is better at one thing, showing sources by default. ChatGPT and Gemini cover more built-in features. See their JAVIS reviews.
Where this came from, and when it was checked.
Access note: perplexity.ai could not be read directly. Perplexity help-center pages were read via Internet Archive snapshots: "Which Perplexity Subscription Plan is right for you?" (Aug 18 update, snapshot Sep 2), "Perplexity Max" (snapshot Aug 26), "How Credits Work on Perplexity" (Aug 6 update, snapshot Sep 6), "What advanced AI models are included" (Sep 4 snapshot), "What is Perplexity Pro?" (snapshot Sep 12), and the Perplexity blog index (snapshot Aug 24). Prices may have changed since.
Pro $20 and Enterprise Max $325: independent trackers (CloudZero, published Sep 21, 2026; GeoToolbox, others) in agreement; not stated in a readable official page.
API pricing: docs.perplexity.ai pricing page, read directly Oct 3, 2026 (no date shown).
Release notes Sept 21: Releasebot aggregation of Perplexity's release notes; GPT-6 Astra in Computer also noted on HN (Sep 6).
Valuation, revenue, traffic share: third-party trackers and Similarweb/FirstPageSage via press; unconfirmed by Perplexity.
Legal: NYT 10-Q and press (NYT suit); CNBC (Mar 10 injunction); Jones Day, Ropes & Gray (Aug 4 Ninth Circuit ruling). Court documents not read directly.
Trustpilot (1.4/5, 825 reviews): direct read of trustpilot.com/review/perplexity.ai, Oct 3, 2026.
Citation audit: Haus Research report (hausresearch.com), Sept 2, 2026, summarised from the report page; HN threads.
Limits complaints and Perplexity statement: Android Authority, May 18-19, 2026 (read directly).
Comet credit-gating: PiunikaWeb, Sep 15, 2026; no official documentation found.
Consumer privacy default: secondary privacy guides (Cape, DeleteMe, others); official policy page blocked.
Official video: youtube.com/watch?v=qOsEYyQuxbU, authorship confirmed on YouTube (author Perplexity, @perplexity-ai).
