Perplexity AI Review 2026: What $20 Actually Buys

Perplexity spent 2026 giving away the thing it used to charge the most for. The Comet browser launched as a perk locked behind the $200-a-month Max plan; it is now free for everyone, on Windows, macOS, Android and iOS. That single move quietly invalidated most of the Perplexity AI review content published before it, because a lot of those articles were selling you a subscription on the strength of a feature you no longer have to pay for.

So this Perplexity AI review deals with what is actually left behind the paywall in August 2026 — model access, search volume, file handling and research quotas — and with the part almost nobody writes about: those quotas have been cut twice without warning, while the price has not moved once. Prices and features below were checked on 19 August 2026 against Perplexity’s own help centre and public reporting. Treat every number as provisional; that is the whole point of this piece.

What Perplexity actually is

Perplexity is an answer engine, not a chatbot with a browser bolted on. You ask a question, it runs searches, reads sources, and returns a short synthesised answer with numbered citations attached to individual sentences. The citation is the product. Everything else — the model picker, Spaces, Labs, Comet — is scaffolding around that one behaviour.

That framing matters because it explains both the appeal and the ceiling. If your work is “find out what is true and where it came from,” Perplexity is very good. If your work is “write me 1,200 words in my voice,” it is the wrong tool and you will be disappointed. We covered that split in more depth in our Perplexity vs ChatGPT comparison for research and fact-checking, and the three-way version in ChatGPT vs Perplexity vs Claude for research.

Perplexity pricing in August 2026

The consumer ladder has been stable on price for over a year. Where third-party trackers disagree is on how many SKUs currently exist — several logged a lineup trim in early July 2026 — so confirm anything unusual on the checkout page before budgeting.

Plan	Price (19 Aug 2026)	Realistically for
Free	$0	Casual, cited replacement for Googling
Pro	$20/mo or $200/yr (≈$16.67/mo)	Daily research users — the tier most people mean
Education Pro	$10/mo, SheerID verification	Verified students and educators
Max	$200/mo or $2,000/yr	People who run agents and Labs all day
Enterprise Pro	≈$40/seat/mo	Teams needing SSO, audit logs, governance
Enterprise Max	≈$325/seat/mo	Orgs running it as core infrastructure

At $20 the Pro plan is price-matched with almost everything else in the category, which we broke down in free vs paid AI tools. The interesting question is not whether $20 is a lot. It is whether the $20 tier still contains anything the free tier does not.

Perplexity AI review: where the $20 genuinely holds up

It does contain real things — four of them.

Model choice under one fee. Perplexity’s help centre lists frontier models from OpenAI, Anthropic and Google available inside a single Pro subscription, alongside its in-house Sonar family. If you would otherwise pay for two assistants, this is the strongest argument on the page. It is also the one you should sanity-check, for reasons in the next section.

Pro Search volume. The free tier caps the deeper multi-step search mode at a handful of runs per day. Pro lifts that to a level most individuals will not hit. This is the upgrade people actually feel within 48 hours.

File and document work. Uploads, Spaces with meaningful file counts, and analysis across a set of documents rather than one at a time. The free tier’s file handling is thin enough to rule it out for anything document-heavy.

Deep Research. The multi-step report mode that reads across a large number of sources and returns something structured. It is the feature Pro subscribers cite most — and, awkwardly, the one that has been cut hardest.

The allowance problem nobody puts in the review

Here is the pattern that should shape your buying decision more than any feature list.

In November 2025, subscribers on Perplexity’s subreddit demonstrated that answers labelled as coming from one model appeared to be served by a cheaper one, verified by inspecting network traffic with a browser extension. The CEO responded publicly and described it as an engineering bug. Whatever the cause, a chunk of the paying community stopped trusting the model label.

Then in February 2026, an announced Deep Research “upgrade” arrived alongside a quota change that users worked out for themselves in the replies: the Pro allowance moved from a very high daily number to a much smaller monthly one. Reporting on community threads puts the drop at roughly two orders of magnitude. It was announced as an improvement, not as a cut.

This is why published limits for Perplexity are all over the place. Different trackers currently report the free tier at three Pro Searches a day or five; Deep Research on Pro at twenty a day or twenty a month. They are not all wrong — they were each correct on the day they checked.

The practical response: while logged in, open perplexity.ai/rest/rate-limit/all in your browser. It returns your account’s live allowances as JSON — Pro Searches, Labs, Deep Research, agentic runs. Check it before you subscribe on the strength of a number in an article, including this one, and check it again after any product announcement. It is the only limit figure that is definitely about your account.

Comet is free now — and that changes the maths

Comet is Perplexity’s Chromium-based browser with an assistant that can read pages, act across tabs, fill forms and complete multi-step tasks. It went free for all users, and mobile builds followed through 2026. It is genuinely useful, and it is no longer a reason to pay.

It is also the part of the product that carries real risk. Agentic browsers are vulnerable to indirect prompt injection: instructions hidden inside a web page, an email, or a calendar invite that the assistant reads as commands from you. Brave’s security team published the first widely-cited demonstration against Comet in 2025, showing it following hidden page instructions and exposing data from a logged-in session. Zenity Labs disclosed a family of zero-click agentic-browser vulnerabilities including Comet in March 2026. Perplexity has shipped fixes; the industry consensus, including from OpenAI’s own commentary on agentic browsing, is that the class of attack cannot currently be eliminated.

Task	Reasonable in Comet?
Summarising articles, comparing pages, research reading	Yes — low stakes, nothing to hijack
Filling public forms, gathering links, drafting text	Usually fine, review before submitting
Acting inside your logged-in email or cloud drive	No — this is the documented attack surface
Banking, health portals, client or payroll systems	No
Autonomous purchasing	Not yet — legally and technically unsettled

The blunt version: use Comet as a reading and research browser, keep a separate ordinary browser for anything with money or credentials in it. That costs you nothing and removes most of the risk.

The legal weather around it

You are not liable for Perplexity’s disputes, but they affect the product you rent. Amazon sued Perplexity in November 2025 over Comet’s shopping agent accessing accounts; a district court granted an injunction in March 2026, and the Ninth Circuit vacated it in August 2026 on the reasoning that the user, not Perplexity, accesses Amazon. The court explicitly limited its ruling to that record. Separately, Perplexity faces a heavy publisher docket including Dow Jones and the New York Post, with other outlets joining through 2026.

The practical consequence is feature volatility. Capabilities that depend on other people’s platforms can disappear on a filing rather than a roadmap. Price that in before you build a workflow on top of one.

Real limitations

Citations are provenance, not verification. A numbered source tells you where a claim came from. It does not tell you the source is reliable, current, or that the sentence summarises it correctly. Misattribution happens. If the answer matters, open the link — the method is in our guide to fact-checking AI-generated content.
Weak at long-form writing. Output is compressed, list-heavy and built for skimming. For drafting you will get better results from a dedicated assistant; see the Claude AI review or the ChatGPT review.
It answers what is indexed and public. Paywalled research, offline records and anything behind a login are simply out of reach unless you upload it yourself.
Allowances move without notice. Twice in four months, subscribers found out about a change by hitting it.
It is an answer engine, not a workflow engine. There is no state, no pipeline, no place for a process to live between sessions.

Who should pay, and who should not

Buy Pro if you research daily for a living — journalism, analysis, due diligence, competitive research, academic work, or client work where the sources matter as much as the answer. At $20 it is straightforwardly worth it, and $10 for verified students is the best value in the category.

Stay on free if you look things up a few times a week, or if what you mostly need is drafting rather than finding. Skip Max unless you are already hitting Labs and agent limits on Pro — the gap between $20 and $200 is the largest unjustified jump in consumer AI unless you are genuinely running agents as infrastructure.

If you are choosing one paid assistant for a small business rather than stacking them, our best AI tools for small business owners roundup makes the case for a different default. And if you have already subscribed and want more from it, start with using Perplexity as a research assistant.

The verdict for this Perplexity AI review: the product is excellent and the pricing is honest, but the thing you are buying is defined by allowances the vendor changes quietly. Subscribe monthly, not annually, until that record improves.

Frequently asked questions

Is Perplexity Pro worth $20 a month in 2026?

Yes if you research daily and value cited, verifiable answers plus access to multiple frontier models under one fee. No if you mainly need writing help, or if you look things up only a few times a week — the free tier covers that comfortably.

Do I need to pay for the Comet browser?

No. Comet launched as a Max-tier exclusive but is now free for all users across desktop and mobile. If a review recommends paying for Perplexity because of Comet, it is out of date.

How many searches do you get on the free plan?

Unlimited basic searches plus a small daily allowance of the deeper Pro Search mode — sources currently report either three or five per day, and the number has changed more than once. Check perplexity.ai/rest/rate-limit/all while logged in for your account’s actual limits.

Is Perplexity Comet safe to use?

It is safe for reading, summarising and research. Independent researchers at Brave and Zenity Labs have demonstrated prompt-injection attacks where hidden instructions on a page or in an email hijack the assistant, so avoid running agentic tasks inside logged-in email, cloud storage, banking or client systems.

Is Perplexity better than ChatGPT?

For sourced research, generally yes — citations are the core product rather than an add-on. For writing, coding and creative work, ChatGPT is the stronger tool. Many people keep the free tier of one and pay for the other.

Can I cancel Perplexity Pro easily?

Yes. Cancel from the subscription section of your account settings on the web, or through the App Store or Google Play if you subscribed through a mobile app. Access continues until the end of the current billing cycle.

Sources

Pricing and feature details checked 19 August 2026. Perplexity changes plan allowances frequently; verify current limits in your own account before subscribing.