Perplexity closed a funding round in January 2026 that valued it at roughly $22.6 billion, more than triple its worth a year earlier, on annualized revenue that crossed $450 million by March. In the same window, its share of daily active users on US mobile did the opposite of what a $22 billion valuation implies: Apptopia data cited across the industry shows it falling from about 6% in October 2025 to roughly 2% by March 2026, a two-thirds drop, while Gemini and Claude gained (Perplexity statistics roundup, 2026).
That gap is the whole story of Perplexity right now. The product that made its name, a search box that cites its sources, is now the least interesting thing it ships. The company spent 2026 turning itself into a browser company and an API vendor, because the thing it invented, cited AI answers, got copied by everyone with a search index. Google’s AI Mode, ChatGPT’s live search, and Gemini’s Android integration all now do the core Perplexity trick natively. So Perplexity moved the fight somewhere Google can’t follow it as easily: the browser itself.
Here is what Perplexity actually is in August 2026, what each piece is for, and where the real risk sits, from the perspective of someone who has to decide whether to let it near a corporate login.
Three products wearing one name
The Perplexity most people picture is the answer engine at perplexity.ai. You ask a question, it searches the live web or a constrained source set, synthesizes a response with a frontier model, and hands you numbered inline citations you can click and verify. That still works, and it is still good at the job it was built for: a fast, sourced answer to a bounded question, the kind of thing that used to cost you ten minutes of reading three articles and triangulating.
But the answer engine is now one of three things under the brand:
- The answer engine (web and app), the original cited-search product.
- Comet, a full Chromium browser with an AI agent built into the sidebar. This is where Perplexity is spending its attention and where its differentiation now lives.
- Sonar, the API that lets developers buy Perplexity’s search-plus-synthesis as a building block, with web grounding and citations included in the token price.
Treating these as one product is how people end up disappointed. The answer engine is a research shortcut. Comet is an agent that acts inside your logged-in sessions. Sonar is infrastructure. They carry very different capabilities and very different risk.
The model layer: still a bet on the interface, not the model
Perplexity never won the model race and never tried to. It bet on the retrieval and interface layer and rented the intelligence, and in 2026 that bet looks smarter than ever because the model field commoditized underneath it.
Two things run the answers. Perplexity’s own Sonar models handle the default, high-volume search traffic: fast, grounded, tuned for citation accuracy rather than raw reasoning. Above that, Pro and Max subscribers get to route a query to the frontier. The current default set inside Comet is GPT-5.6 Sol, Claude Opus 5, and Gemini 3.1 Pro, each with extended thinking, and you can swap in others, including Grok 4.5, per query.
This is the same routing logic I have argued for repeatedly on this site: match the model to the workload instead of defaulting to the most expensive option. Perplexity bakes that decision into the product. If you already understand why every frontier model now ships an effort dial and how to build a cost map instead of chasing a cheaper model, Perplexity’s model picker is just that idea with a nicer interface. For a fuller read on the underlying models, the site’s Claude, Gemini, and Grok guides go deeper on each.
One number worth keeping: on the Columbia Journalism Review’s citation-accuracy testing, Perplexity’s Sonar Pro posted the lowest citation-hallucination rate among the major AI search platforms, a meaningful lead over ChatGPT Search and a large one over Grok. That is the metric that actually matters for a cited-answer product, and it is the one place Perplexity still clearly leads. A tool that cites confidently but fabricates the citation is worse than useless, and Perplexity fabricates less.
Comet is the real 2026 story
Comet launched in mid-2025 locked behind the $200-a-month Max tier. Four months later, on October 2, 2025, Perplexity made it free worldwide (CNBC). By 2026 it runs free on macOS, Windows, Android, and iOS, with no account required to download and browse.
The pitch is not a browser with a chatbot bolted on. Every new tab opens with the Comet Assistant in the sidebar, and it is an agent, not a search box. It reads the page you are on, holds context across your open tabs, and can take actions: summarize a long thread, pull the key terms out of a contract, draft a reply, compare products across tabs, walk through a checkout. It browses on your behalf inside your authenticated sessions, which is exactly what makes it useful and exactly what makes it dangerous.
Perplexity also built a revenue layer on top of it. Comet Plus, $5 a month and included free with Pro and Max, buys access to premium publisher content from partners including CNN, The Washington Post, Fortune, the Los Angeles Times, and Condé Nast titles such as Wired, The New Yorker, and Vogue (Perplexity, Comet Plus launch partners). It is Perplexity’s answer to the publisher backlash over uncompensated scraping: pay the sources, then resell their content inside the browser. For a deeper walk-through of the browser itself, see the site’s Comet coverage.
The security asterisk you cannot skip
I spent twenty years in IT operations, and the moment a tool asks to act inside a credentialed session, my first question is not “how smart is it” but “what happens when a web page tells it to do something I didn’t.” For Comet, the answer in 2026 is: quite a lot.
Comet has been the single most-breached AI browser of the year, and the flaws are not bugs you patch once. In October 2025, researchers at LayerX disclosed CometJacking, an attack that hides instructions in URL parameters so a single malicious link can push Comet to reach into connected services such as email and calendar and exfiltrate data, with no credentials entered and no further clicks (The Hacker News). In 2026, Zenity Labs disclosed a family it called PleaseFix / PerplexedBrowser: zero-click prompt injection seeded inside an ordinary calendar invite that could exfiltrate local files or hijack a user’s 1Password session if the extension was installed and unlocked (Zenity Labs).
The common thread is architectural, not incidental. Comet mixes three things that should never share a trust boundary: untrusted content from the open web, a credentialed session, and the authority to act. Security researcher Simon Willison named that combination the “lethal trifecta,” and the UK’s National Cyber Security Centre has said indirect prompt injection may never be fully eliminated. This is not a Perplexity-only failing. I walked through the whole field, Comet, ChatGPT Atlas, Claude for Chrome, and the rest, in the AI browser security piece from earlier this month; Comet just happens to sit at the aggressive end of the spectrum, agent-on by default across every platform.
None of that means don’t use Comet. It means use it the way you would provision a privileged service account. Run it in a dedicated profile that is not logged into your bank, your password manager, or your primary email. Keep read-only browsing separate from anything that can send, spend, or change state. Require a human approval step for consequential actions. Treat every page it reads as potentially hostile input, because to the model it is indistinguishable from your instructions.
Pricing in 2026: what each tier is actually for
Rates move constantly, so verify at perplexity.ai before you commit. As of August 2026 the structure is:
| Tier | Price | What you get | Who it’s for |
|---|---|---|---|
| Free | $0 | Sonar-powered cited search, limited deep-research runs, Comet browser free on all platforms | Casual users, anyone testing it |
| Pro | ~$20/mo ($200/yr) | Unlimited Pro Search, frontier model routing (GPT-5.6 Sol, Claude Opus 5, Gemini 3.1 Pro, Grok 4.5), Spaces, file uploads, Comet Plus included | Researchers, knowledge workers, power users |
| Max | $200/mo ($2,000/yr) | Everything in Pro plus highest usage ceilings, earliest feature access, priority compute | Heavy daily users, professionals billing the time back |
| Enterprise | Custom | Team Spaces, SSO, data controls, admin, API access | Organizations with compliance requirements |
For developers, Sonar is priced separately from the consumer tiers. The base Sonar model runs around $1 per million input and output tokens; Sonar Pro is roughly $3 per million input and $15 per million output, with a 200K-token context window, plus a per-request search fee that scales with search depth (CloudZero, Perplexity API pricing). The important detail: web search and citations are included in that price, with no separate search-provider fee. If you are building a product that needs grounded, cited answers and you don’t want to assemble your own retrieval pipeline, Sonar is often cheaper than bolting search onto a raw frontier model. If you are just running a lot of well-scoped queries, price it against self-hosting first.
The free tier remains genuinely usable, more so than most freemium AI products. If you run more than a handful of research queries a day, Pro pays for itself in time saved. Max is a narrow product: it makes sense only if you are a heavy daily user who bills the hours back, and most people who think they need it actually want Pro.
Where it wins, where it doesn’t
The honest 2026 read:
- Fast, cited answers from the live web. Still best-in-class, and the citation-accuracy edge is real. This is the core job and Perplexity does it better than the general chatbots.
- Research you need to trust and share. Spaces (persistent, source-constrained research environments) and Pages (turning a thread into a shareable brief) are quietly the most useful features for teams that want a lightweight substitute for building their own retrieval setup.
- Agentic browsing inside your sessions. Comet is capable and genuinely convenient, with the security caveats above. Convenience and blast radius are the same feature here.
- Deep reasoning over long documents. Weaker. Reach for Claude or GPT directly, or for NotebookLM when the whole job is grounded synthesis over a fixed corpus.
- Mobile and OS-level assistance. This is the soft spot the DAU numbers expose. Gemini’s Android integration and ChatGPT’s reach mean Perplexity has no home-field advantage on a phone, and it is losing share there fastest.
The strategic question hanging over all of it: Perplexity invented a category and then watched the platform owners absorb it. Its bet is that a dedicated, model-agnostic browser and a clean search API are defensible in a way a cited-answer box is not. The valuation says investors believe it. The mobile engagement numbers say the market hasn’t decided. For where Perplexity sits against the rest of the field, the site’s platform landscape guide maps the full 2026 lineup.
The practitioner takeaway is simpler than the strategy. Use the answer engine as a research accelerator and trust its citations more than its rivals’. Use Sonar if you are building and want grounding without the plumbing. Use Comet if the convenience is worth standing up a sandboxed profile to contain it, and do not skip that step, because the browser that acts for you is the same one an attacker can point a hidden instruction at.