Perplexity in 2026: $450M ARR, a 37% citation-failure rate, the free Comet browser, and eight publisher lawsuits β what the numbers actually say.
Perplexity is still the best tool for one job: cited, current answers. That job got materially bigger this year β Comet went free worldwide and Deep Research learned to output decks, spreadsheets and dashboards instead of just prose. Rating: 4.3/5.
3.6/5. Generation, memory and ecosystem still belong to the general assistants. Perplexity is a research front-end, not a ChatGPT replacement, and treating it as one produces a frustrating experience for reasons that have nothing to do with quality.
Three things: the browser stopped costing $200/month and became free; revenue more than doubled; and the legal exposure went from "a nuisance" to "a structural risk with eight named plaintiffs." All three matter to a buying decision.
The product that defined the category: web-searching, source-citing answers with inline attribution. Everything else the company has built sits on top of this loop, and it remains the fastest path from "I need to know X as of today" to a linked answer.
Autonomous multi-source investigation β 20 runs/day on Pro β now generating presentations, spreadsheets, dashboards and websites rather than only reports. It runs on frontier models (Claude Opus-class for Pro/Max tiers). The output-format expansion is the more important half: a cited report that lands as a deck is materially closer to finished work than one that lands as markdown.
The agentic browser: free on iOS, Android, Mac and Windows since March 18, 2026. It reached #3 on the US App Store within 48 hours of the iOS launch. Roughly 3 million monthly active users β meaningful, but a small fraction of Perplexity's overall base, which tells you the browser is still an early-adopter product rather than the default surface.
MDM-deployable Comet with granular agent-permission policies, starting around $40/seat. The permission model is the actual product here β an agentic browser inside a regulated environment is a governance problem before it is a capability question.
ARR crossed $450 million in March 2026, up from roughly $200 million in September 2025 β more than doubling in six months. Third-party projections put full-year 2026 revenue near $656 million. Whatever you think of the product, that is not the growth curve of a feature; it is the growth curve of a business finding its market.
This is worth flagging because the discrepancy is large:
| Figure | Source framing |
|---|---|
| ~45 million MAU | Core product, mid-2026 (up from 22M at start of 2025) |
| ~100 million MAU | "Across app and search" β a broader surface definition |
| ~3 million MAU | Comet browser specifically |
The 45M and 100M figures are not reconcilable without knowing exactly what "search" includes. Use 45M when comparing to other assistants' core products; treat the 100M number as marketing-adjacent until someone defines the denominator.
Perplexity raised about $200 million at a valuation near $20 billion in June 2026, bringing total funding to roughly $1.72 billion. Some coverage cites a $23 billion figure from January 2026. Both numbers circulate; the January figure appears to be a different round or a secondary. The honest summary: low-to-mid twenties of billions, on under half a billion of ARR.
Columbia Journalism Review's Tow Center audit remains the most-cited independent check, and the results are unflattering across the board:
| Platform | Citation-failure rate |
|---|---|
| Perplexity (Sonar Pro) | 37% β lowest measured |
| ChatGPT Search | 67% |
| Grok 3 | 94% |
Two honest readings coexist: best in class by a wide margin, and still wrong on more than one citation in three. Both are true. Neither cancels the other.
A Towards AI analysis found 92.3% citation accuracy for Perplexity Deep Research versus 87.6% for ChatGPT's equivalent. That gap is real but the source is a blog post, not a peer-reviewed study β treat it as directional evidence, not as a measurement.
Why the two numbers differ so much: the CJR audit tests fast, single-pass answers; Deep Research runs a multi-step verification loop. The mode you use changes your error rate more than the vendor you choose does.
Frontier-model hallucination rates in 2026 sit between 3% and 19% depending on task, down from 15β45% in 2024. Citation attribution has improved far more slowly than factual accuracy β which is the specific failure mode that matters for a research tool.
Treat citations as leads, not as verification. Click through before you quote. With Perplexity you will be right more often than with any competitor, which is the correct bar for a research tool β not the bar for a truth oracle. (How it fits a rigorous stack β)
| Date | Comet availability |
|---|---|
| July 2025 | Desktop only, $200/month Max subscribers |
| October 2, 2025 | Free worldwide, desktop |
| March 18, 2026 | Free on iOS, Android, Windows, Mac |
A company does not go from $200/month to free in eight months because the product got worse. It does so because it decided distribution matters more than the revenue that tier was producing.
The agentic model means Comet doesn't just surface links β it navigates, summarizes, collates across tabs, and completes multi-step user-initiated tasks. The assistant side panel that reads the current page is the feature most users actually adopt; the multi-tab automation is the one demos lead with.
Agentic browsing is still the least reliable thing in the product. Tasks that touch authentication, dynamic forms or checkout flows fail more often than they succeed, and the failure mode is usually silent β the agent reports completion for something it didn't finish. Use it for reading and synthesis; supervise it for anything that writes.
For anything that happened this week, Perplexity beats the chat assistants' browsing modes on both speed and source quality. This remains the single clearest capability gap in its favor.
The workflow starts verifiable instead of requiring a "cite your sources" prompt that models comply with unevenly. Small design decision, large practical difference.
Summarize-this-page, compare-across-tabs, one-tap follow-ups β research woven into browsing rather than bolted onto a chat window. Free, which removes the main objection to trying it.
The deck/spreadsheet/dashboard output is the most underrated feature shipped this year. It moves the tool from "produces a thing you then have to reformat" to "produces the thing."
Drafting, rewriting, coding β mediocre next to Claude or ChatGPT. This is a positioning choice, not a defect, but it's the reason "just use Perplexity for everything" fails.
Nothing comparable to ChatGPT's persistent memory. Every session starts cold. For repeated research on the same topic, you re-establish context every time β a real tax on long-running projects.
No meaningful image generation, thinner third-party integrations, fewer places the tool shows up in an existing workflow. Distribution via the browser is the company's answer to this, and it's early.
The multi-vendor model menu rotates fast. Aggregate capability is stable; specific model names are not. Don't build documentation or training material around a named backing model.
Full Comet plus core answers. Genuinely substantial since March, and now the most generous free tier in AI search by a distance. (Free-stack context β)
$16.67/month billed annually. Deep Research allowance, frontier-model backing, file analysis, and Comet Plus bundled in. The correct buy for anyone doing research weekly rather than occasionally.
Model choice for the browser agent, top limits, earliest feature access. Power-user territory; hard to justify unless the agentic browser is central to your daily work.
The governance layer: MDM deployment, agent-permission policies, audit surface. Priced per seat with contact-required tiers above the entry point.
Unlocks premium partner-publisher content inside the browser. Bundled into Pro and Max; available standalone. 80% of that fee goes to participating publishers β which is where the pricing story connects to the legal one.
The plaintiff list is long and growing: Dow Jones (WSJ), The New York Times, Forbes, CondΓ© Nast, and Japan's Nikkei and Asahi Shimbun. CNN filed a federal copyright and trademark suit alleging Perplexity scraped more than 17,000 CNN stories, videos and images without authorization.
Perplexity's counter-move is a $42.5 million revenue-share fund and the Comet Plus 80% split, with participating outlets including Gannett, TIME, Fortune and the Los Angeles Times. Whether this reads as good-faith partnership or legal risk mitigation depends largely on which side of a lawsuit you're on.
Litigation risk rarely kills a product outright, but it does shape one. The realistic outcomes are licensing costs that pressure margins, or content availability that varies by publisher. Neither is a reason to avoid the tool today; both are reasons to avoid making it a single point of failure in a research workflow you can't reconstruct elsewhere.
Start every "what's true about X right now" question here. Take the citations, click through, and carry the sources β not the summary β into whatever tool you write in. This is the workflow the product is genuinely best at.
Perplexity for finding and verifying; a generalist for drafting and reasoning over what you found. The 45M-user base is largely people who arrived at this arrangement independently.
Given a 37% citation-failure rate, the discipline isn't optional. Rule of thumb: anything you'd be embarrassed to be wrong about in public gets clicked through. Anything you're using to orient yourself doesn't. Most research is the second category, which is why the tool works despite the number.
Perplexity spent 2026 converting a feature β "search with citations" β into a platform: browser, research engine, enterprise governance layer. The free Comet move buys reach; Deep Research artifacts buy stickiness; the publisher fund buys time. It remains a complement to a general assistant rather than a substitute. But in its lane, with a 37% citation-failure rate against competitors at 67% and 94%, nothing else is close.
Sources: DemandSage β Perplexity AI statistics 2026 Β· ValueAdd VC β valuation and ARR analysis Β· TechTimes β the $200M Comet raise Β· Suprmind β AI hallucination rates and benchmarks Β· eesel β Comet pricing breakdown Β· Axios β publishers get 80% of Comet Plus revenue Β· Accelerate IP β CNN v. Perplexity
Last updated: July 30, 2026
Seedance 2.5 brings 30-second audio-video generation, 50 mixed references and timed editing. Here is what is official, priced and still worth testing.
MiniMax H3 generates 2K video with native stereo audio, publishes per-second pricing, and offers H3-Base weights under a Community License.
Hermes Agent blends persistent memory with self-improving skills, but its security and outcome depend on the backend, approvals, and review you choose.
OpenClaw is a high-capability agent runtime whose value depends on deliberate gateway security, skill review, and operating discipline.
AI music in 2026: Suno v5.5's capabilities and legal defiance, Udio's settlement path, the Warner deal retiring unlicensed models, and creator rules.
The AI video market after Sora's exit β with real pricing: $0.50 to $2.50 per 10-second clip, who leads the rankings now, and how to choose per shot.
The honest 2026 free-tier comparison: Gemini's generous bundle, Claude's quality-first plan, ChatGPT's capped breadth, and when free stops being enough.
Archived review of GPT-5's August 2025 launch: the unified-router design, the 4o backlash, what improved β and how it reads from mid-2026.
