Perplexity

Perplexity now runs two currencies, and the distinction decides what a plan is worth to you.

Perplexity's help centre states plainly: "Perplexity Ask searches do not consume credits." Credits are consumed only by Computer, the agent that carries out multi-step work — research, file and app creation, connector-driven tasks, scheduled jobs (https://www.perplexity.ai/help-center/en/articles/13838041-how-credits-work-on-perplexity).

The same page sets the exchange rate: "Today, 100 credits equals $1." And it gives the vendor's own per-task bands:

• Light (research threads, summarisation): 100–350 credits (~$1–3.50)
• Complex (multi-step reports, analysis): 350–950 credits (~$3.50–9.50)
• Heavy (complex documents, research-and-code): 875–2,275 credits (~$8.75–22.75)
• Mega (large projects, video, data collection): 2,400–9,800 credits (~$24–98)

What each plan includes, in the vendor's words: "Consumer Pro plans do not start with a set monthly amount. Consumer Max plans start with 10,000 credits a month." And more bluntly, further down the same page: "After your bonus runs out, Pro plans do not include monthly credits. To keep using Computer as a Pro subscriber, you buy credits."

So a $20/month Pro subscriber gets unlimited-ish search and effectively pay-as-you-go agent use. A $200/month Max subscriber gets 10,000 credits — roughly $100 of credits at the stated rate — which against the vendor's own bands works out to something like 10–28 "Complex" tasks, 4–11 "Heavy" ones, or 1–4 "Mega" ones per month. Unused monthly credits "expire at the end of each billing cycle"; purchased credits expire after a year of inactivity; and on cancellation "all credits, monthly and purchased, are forfeited".

Two guardrails worth knowing: auto-refill is off by default and Perplexity states "You'll never be automatically charged for credits without opting in first", and every Max account has a monthly spending cap defaulting to $200 that you can raise "anywhere up to $5,000". If a task fails on Perplexity's side the credits are restored; if it finishes but the result is poor, "credits are not automatically refunded".

One operational trap from the same page: if you bought Max through the iOS or Android app, "you'll need to add a card on the website to use Computer."

The strongest independent evidence is the EBU/BBC study News Integrity in AI Assistants (October 2025), covering 22 public service media organisations across 18 countries and 14 languages, with over 3,000 responses assessed by professional journalists (https://www.ebu.ch/research/open/report/news-integrity-in-ai-assistants).

Perplexity came out best of the four assistants tested. From the report: "Gemini recorded the highest proportion of significant issues, impacting 76% of responses. This was double the rate of the next assistant Copilot (37%), followed by ChatGPT (36%) and Perplexity (30%)." On sourcing specifically — the study's single biggest cause of significant issues — "Gemini had significant issues in 72% of responses. This was three times as many as ChatGPT (24%), followed by Perplexity and Copilot (both 15%)."

That is a genuine, well-measured technical strength, and it is the thing Perplexity is actually built around. If citation quality is why you are considering the product, the independent evidence supports the choice.

The same study is equally specific about where Perplexity still fails, and a buyer should weigh both. Assessors recorded that "Perplexity (BBC custom) fabricated or altered three separate quotes", including a quote attributed to the Unite union general secretary that "is not in the sources cited". A response for German broadcaster ARD claimed Tagesschau "describes Orbán's rule as authoritarian and illiberal" — a characterisation "not found in the cited Tagesschau source". And the report notes a padding pattern in the citation lists that are Perplexity's selling point: "Perplexity (NRK) appended a sources block with 19 URLs but only referenced three of the sources in the body of the answer."

The headline for the whole study, across all four assistants: "45% of responses contained at least one significant issue of any type." Best-in-class here still means roughly three in ten news answers carry a significant problem.

Context on age: this is an October 2025 study and the most recent large-scale multi-assistant evaluation we could verify as of August 2026. An earlier Columbia Journalism Review / Tow Center study (March 2025) found something worth pairing with it — that paid tiers were not simply better: "our tests showed that while both answered more prompts correctly than their corresponding free equivalents, they paradoxically also demonstrated higher error rates", because the paid tiers answered confidently where free tiers declined (https://www.cjr.org/tow_center/we-compared-eight-ai-search-engines-theyre-all-bad-at-citing-news.php). Per-tool percentage figures from that CJR study circulate widely (commonly "37%"); we could not verify them in the article text and are not repeating them.

Perplexity publishes a plan comparison table (https://www.perplexity.ai/help-center/en/articles/11187416-which-perplexity-subscription-plan-is-right-for-you), and the notable thing is which numbers appear and which don't.

Free is quantified: 3 Pro Searches per day, 1 Research query per month, no Comet Assistant, no advanced models, no image generation.

Enterprise is quantified. Enterprise Pro: 400 Pro Searches/week, 50 Research queries/month, 80 Comet Assistant queries/month, 50 file/app creations/month, 100 upload sessions/week. Enterprise Max: 4,000 Pro Searches/week, 500 Research/month, 800 Comet Assistant/month, 500 file/app creations/month, 1,000 upload sessions/week. The enterprise billing FAQ adds that Enterprise Max supports "up to 10,000 personal files and 5,000 files per project, and up to 15 video generations each month" (https://www.perplexity.ai/help-center/en/articles/10352986-enterprise-pricing-and-billing-frequently-asked-questions).

Consumer Pro and Max are not quantified anywhere we could find. The table describes them only as "Weekly limits (average use)" for Pro and "Weekly limits (advanced use)" for Max. If you are an individual trying to work out whether $20 or $200 buys enough headroom for your workload, the vendor does not publish the number, and we could not establish it. Note that Enterprise Pro's disclosed 400 Pro Searches/week is described as "extended usage limits compared to Perplexity Pro", which bounds consumer Pro from above but doesn't pin it down.

One pricing detail confirmed by the vendor: "Max annual billing is only available on perplexity.ai" — the app stores sell Max monthly only.

Perplexity's data collection page, last updated 16 July 2026, is unambiguous: "For Free, Perplexity Pro and Perplexity Max users, AI Data Retention is enabled by default." You can opt out in account settings, but you have to go and do it (https://www.perplexity.ai/help-center/en/articles/11564572-data-collection-at-perplexity).

For Enterprise the position reverses: "As a Perplexity Enterprise user, your data is never used for AI training purposes", and that holds even if the subscription lapses into past-due status.

The transition is the part people miss. Per the same page, if an organisation downgrades from Enterprise to a Free or consumer plan, "your data would then follow the standard consumer data policy, where AI data retention is enabled by default. In that case, you'll need to manually opt out." A team that lets an Enterprise contract lapse into individual Pro seats silently re-enables training unless each person changes the setting.

The plan comparison table frames the same split as "Opt-out AI training available" for Free/Pro/Max versus "Data never used for model training" for both Enterprise tiers.

Perplexity states on its enterprise page that it is "SOC 2 Type II Certified" and "GDPR and HIPAA compliant", offers SSO and SCIM, and supports "private search and automatic file deletion up to 1 day" (https://www.perplexity.ai/enterprise). We have not seen the audit report; treat the certifications as vendor claims pending your own diligence, which for SOC 2 means requesting the Type II report under NDA.

From Perplexity's enterprise billing FAQ (https://www.perplexity.ai/help-center/en/articles/10352986-enterprise-pricing-and-billing-frequently-asked-questions):

• No trials. "We do not offer trials for Enterprise Pro or Enterprise Max."
• The API is not included. "No, API usage is not included with Enterprise Pro or Enterprise Max. Users who need API access must purchase credits or top off their balance with a valid payment method. Note that administrators cannot manage API credits for other users." API credits are also "billed separately" from the seat invoice. If your plan was to give the team an enterprise seat and let engineers hit Sonar from it, that doesn't work.
• 30-day seat commitment. "During your first month of subscription, you'll be charged for the number of seats you selected during onboarding. Because there's a 30-day seat commitment period, any seat removals will only show up in your billing after the first month." Later changes prorate; billing runs weekly for monthly subscriptions.
• Mixed tiers are allowed. "Organizations can flexibly combine Enterprise Pro and Enterprise Max seats" — so you can put a handful of heavy researchers on Max without moving everyone.
• Invoice billing is annual-subscription only; wire transfer is available at checkout; white-glove invoicing at 250+ users.

On the list price: the FAQ says "We offer two pricing tiers" and then renders the figures in a client-side component that we could not extract, and www.perplexity.ai is behind a Cloudflare challenge that blocked both curl and a real browser. So the Enterprise Pro and Enterprise Max list prices are not established here. Figures of $40 and $325 per seat per month circulate on third-party pricing blogs; none of them cite a primary source and we are not repeating them as fact.

One per-seat number we could verify from Perplexity directly: "We offer Enterprise Pro at a discount of $30/seat per month or $300/seat annually for eligible educational institutions (not covered by our standard Student Program) and nonprofit organizations" — application required, and per the billing FAQ "Students cannot request this discount directly. An IT representative or department head must contact us."

Consumer prices we did verify, from Perplexity's own App Store listing (https://apps.apple.com/us/app/perplexity-ai-search-chat/id1668000334): Perplexity Pro at $20.00 and $200.00, Perplexity Max at $200.00, plus credit packs at $5.00/500, $10.00/1,000, $25.00/2,500, $50.00/5,000 and $100.00/10,000 — consistent with the vendor's 100 credits = $1.

If you're buying Perplexity as infrastructure rather than as a subscription, note that the API bills on two axes at once — tokens and per-request search fees. Verified today against the raw pricing table (https://docs.perplexity.ai/docs/getting-started/pricing):

On top of tokens, Sonar Pro and Sonar Reasoning Pro each carry request fees of $6 / $10 / $14 per 1,000 requests for low / medium / high search context size. Pro Search mode raises that again: the pro setting ("multi-step tool usage for complex queries") is priced at $14 / $18 / $22 per 1,000 requests, against $6 / $10 / $14 for default Sonar Pro behaviour. At high volume the request fee, not the token spend, is likely to be the line item that surprises you — model your traffic in requests before you model it in tokens.

Throughput is gated by cumulative spend rather than negotiated up front (https://docs.perplexity.ai/docs/admin/rate-limits-usage-tiers):

Two things follow. A new account starts at 1 QPS, so load-test before you assume a launch is viable. And the ceiling is 33 QPS — Tier 5 buys no more throughput than Tier 4, so if you need more than that, it's a sales conversation, not a spend threshold.

We did not establish the API's data retention or training policy, or any uptime SLA. Ask before you send customer data through it.

Perplexity's answers are only as good as the sites it can reach, so its running disputes with publishers and platforms are a product question, not just a reputational one.

Amazon. Amazon sued over Comet's agentic shopping and won a preliminary injunction in March 2026. On appeal the Ninth Circuit vacated it — reported 4 August 2026 — reasoning that because Comet acts on user direction, it is the user rather than Perplexity accessing Amazon's servers, so Amazon was unlikely to succeed on its Computer Fraud and Abuse Act claim; the court added that an injunction "against conduct that likely does not violate the CFAA or the CDAFA would not serve the public interest" (https://www.engadget.com/2230471/perplexity-has-successfully-overturned-amazon-injunction-on-its-ai-shopping-bot/). Amazon told Engadget it "respectfully disagree[s]" and is "evaluating our next steps". This is a meaningful win for anyone relying on agentic browsing, but the underlying case continues in San Francisco federal court and rehearing or Supreme Court review remain open.

Dow Jones / NY Post. The copyright and trademark suit filed in 2024 is still live and still in discovery. We checked the docket directly: fact discovery running into mid-2026, expert discovery to 19 October 2026, pretrial conference adjourned to 14 December 2026, and the parties estimate a 10–14 day jury trial (https://www.courtlistener.com/docket/69280523/dow-jones-company-inc-v-perplexity-ai-inc/). No ruling on the merits yet — nothing here is settled either way.

Cloudflare. In August 2025 Cloudflare published research alleging Perplexity used "stealth, undeclared crawlers", saying it observed Perplexity "repeatedly modifying their user agent and changing their source ASNs to hide their crawling activity", and de-listed Perplexity as a verified bot (https://blog.cloudflare.com/perplexity-is-using-stealth-undeclared-crawlers-to-evade-website-no-crawl-directives/). Perplexity rejected this in detail, arguing Cloudflare "confused Perplexity with 3-6M daily requests of unrelated traffic from BrowserBase, a third-party cloud browser service that Perplexity only occasionally uses for highly specialized tasks (less than 45,000 daily requests)", and that user-driven agents are categorically different from crawlers because content is "used immediately to answer your question" and not stored or trained on (https://www.perplexity.ai/hub/blog/agents-or-bots-making-sense-of-ai-on-the-open-web). Both posts date from August 2025 and we could not establish the current verified-bot status; the underlying disagreement about whether user-triggered agents should obey robots.txt is unresolved industry-wide, not specific to Perplexity.

The practical read for a buyer: coverage of specific publishers may vary and may change without notice. The EBU study independently observed related oddities — that assistants sometimes declined or got wrong questions about publishers that did permit access, and sometimes answered correctly about publishers whose content they shouldn't have had.

Vendor-continuity risk looks low. The most recent round we could verify: TechCrunch reported in September 2025 that Perplexity raised "$200 million in new capital at a $20 billion valuation", citing The Information, following "$100 million at an $18 billion valuation" in July 2025 and a $500 million round at $14 billion led by Accel earlier that year, with "$1.5 billion in total funding" per PitchBook (https://techcrunch.com/2025/09/10/perplexity-reportedly-raised-200m-at-20b-valuation/). ARR at that point was reported as "approaching $200 million" by a source familiar with the company, against "more than $150 million" stated by Perplexity's own head of communications. Higher 2026 valuation and ARR figures circulate on secondary-market and SEO sites; we could not verify any of them against a primary source and are not repeating them.

CNBC's May 2026 Disruptor 50 profile reports Perplexity's search engine "answers 1.5 billion questions worldwide monthly", that Comet gained a free tier in October 2025, and that the company has launched "Perplexity for Gov" with the U.S. General Services Administration alongside FedRAMP compliance work (https://www.cnbc.com/2026/05/19/perplexity-cnbc-disruptor-50-ranking.html). Disruptor 50 metrics are company-supplied, so treat the usage figure as a vendor claim reported by CNBC rather than an audited number.

Adoption claims to discount accordingly: CNBC's profile relays that Perplexity "counts 92% of Fortune 500 companies as users" (a user-presence figure, not a paid-contract figure), and the enterprise page claims "50,000+ organizations trust Perplexity Enterprise". Neither is independently corroborated.