Is Perplexity Down Right Now? 2026 Status Check

I checked is perplexity down right now against the official status page, third-party monitors, and the failure patterns I see most often in production AI workflows, and the answer is no: Perplexity is operational, even though a slow response, quota wall, cached session, connector fault, or regional network route can still make the service feel broken. That contradiction is the useful part. A tool can be up globally while a user in London, Karachi, New York, or Dubai sees a frozen browser tab, a failed file upload, or an API timeout.

This article separates the live service question from the user experience question. I start with the operational evidence, then move into the common reasons Perplexity appears unavailable when the platform itself is still online. That includes browser cache, authentication state, VPN routing, Pro Search quota, Research limits, connector errors, API billing, token costs, and model latency. I also explain why one tracker may say degraded performance while another says all clear, because outage monitors are not all measuring the same thing.

During our 2026 evaluation, I treated Perplexity as five connected systems rather than one website: public web app, logged-in workspace, mobile and Comet surface, connectors, and API. That view gives a cleaner answer than refreshing a single page. It also helps teams avoid over-escalation. A real incident requires official evidence, repeated failures across networks, and failed authenticated and unauthenticated checks. A local issue usually disappears after a targeted cache, account, network, or quota fix.

Is Perplexity Down Right Now? The Live Evidence

The strongest answer comes from the official status page, because it reports platform components instead of only collecting public complaints. At the time of verification, the Website and API components were marked operational, recent uptime for core components was shown at or near 100 percent, and the notice history displayed no active notices in the previous seven days. That makes the current evidence point away from a platform-wide Perplexity outage.

Third-party monitors add context, but they should not outrank the vendor status feed. Entireweb reported Perplexity operating normally on 21 June 2026, with 0 user reports in the previous 24 hours and a response time of 293.4 ms. DownForEveryone Or Just Me reported no current problems and listed the previous detected outage as 3 June. StatusGator also showed the official component as operational, although community-report panels can still display local complaints or inconsistent counts. I treat those discrepancies as a reminder to separate server state from user state.

Source	Signal observed	How to interpret it
Official status page	Website and API operational, with no recent active notice in the checked window.	Highest-priority source for platform health because it comes from Perplexity components.
Entireweb status monitor	Operating normally, 0 recent reports, and 293.4 ms response time.	Useful public confirmation that the main website was reachable from a monitoring perspective.
DownForEveryone Or Just Me	No current problems detected, with a previous outage listed on 3 June.	Useful for quick public reachability, but not a full product-state diagnostic.
StatusGator	Official status feed operational, with community-report data that can vary.	Good for change alerts, but user reports may reflect local errors or regional routing.
Your browser or API client	May still fail while monitors look clean.	Treat as a local diagnostic problem until multiple independent checks fail.

The practical verdict is clear: no broad outage is visible from the strongest sources. If Perplexity is not working for you, the next step is not to wait for a global fix. The next step is to prove whether the failure follows your account, browser, device, network, workspace, or API key.

What Operational, Degraded Performance and Down Mean

Outage language is messy because status tools compress very different failure modes into a few labels. Operational means the monitored component is responding within the vendor’s acceptable range. Degraded performance means the service is still responding, but latency, error rate, or a sub-component may be worse than normal. Down means a component is unavailable or failing widely enough to merit an incident. To understand the product surface behind those labels, it helps to review the feature architecture behind Perplexity, because search, citation retrieval, Research, file understanding, and model selection do not all fail in the same way.

Is Perplexity Down Right Now or Just Slow?

The most common false alarm is a slow response that gets described as downtime. A long wait on a Deep Research query can feel like a failure, but it may be the result of multi-step retrieval, reranking, citation processing, tool calls, or a larger context window. In our hands-on testing, standard search checks were useful for availability, while Research and file upload checks were better for stress testing slower workflows. The two measurements should not be mixed.

The right mental model is a layered stack. The page shell can load while the model response stalls. Login can work while a connector fails. The API can answer a simple request while a larger context request times out. A mobile app can fail because of a stale token while the desktop site works normally. If a monitor says degraded performance, do not assume Perplexity is offline. Assume a measurable part of the experience is slower or less reliable than the baseline.

For newsroom, research, and engineering teams, the threshold for declaring an incident should be stricter than one slow prompt. I use three tests: an unauthenticated homepage check, an authenticated account check, and an API or workflow-specific check if the team depends on integrations. When all three fail from more than one network, the probability of a real incident rises sharply. When only one layer fails, local troubleshooting is usually faster than switching tools.

The Fast Diagnostic Ladder for Perplexity Connection Errors

When Perplexity looks unavailable, start with the lowest-cost tests. The point is not to perform every fix randomly. The point is to isolate the layer that changed. The Perplexity troubleshooting guide is useful because it treats loading loops, extension conflicts, cookies, stale sessions, and account state as separate failure classes rather than one vague complaint.

1.  Open the official status page from a normal browser tab and confirm whether Website, API, and related components show operational or an active incident.

2.  Open Perplexity in a private window without extensions. If it loads there, suspect cache, cookies, an extension, or a stale login token.

3.  Switch networks, such as Wi-Fi to mobile hotspot. If the problem vanishes, suspect DNS, VPN, firewall, ISP routing, or a corporate gateway.

4.  Sign out and sign back in, then verify whether the same account can run a simple search before testing file upload or Research.

5.  Check quota and billing state before escalating. A capped plan, expired subscription, failed payment, or exhausted Research allowance can look like a broken product.

6.  For API users, test the smallest possible request with the same key, then test a larger request. If only the large request fails, investigate context size, timeout, streaming, or tool-call settings.

This ladder keeps the diagnosis narrow. If a private window works, do not blame the service. If another network works, do not clear your entire workspace. If a simple API call works but a tool-heavy request fails, the core endpoint is not necessarily down. The best troubleshooting sequence reduces uncertainty with the fewest account and environment changes.

A common mistake is clearing everything before preserving evidence. If you are diagnosing an enterprise problem, capture the time, workspace, browser version, network, request type, error message, and whether the failure affected only one user or many. That small incident note is more useful to support than a vague statement that Perplexity is down.

Local Failure Modes That Mimic an Outage

Most Perplexity availability complaints I see are not platform outages. They are local failures with outage-like symptoms. A blank page can be an extension conflict. A repeated login prompt can be a cookie issue. A file upload failure can be a format, size, or workspace quota problem. A Research button that seems inactive can be a monthly limit rather than a server incident.

The distinction matters because each symptom has a different fix. Clearing cache helps a stale browser session, but it does nothing for a corporate firewall. Changing networks helps routing, but it does nothing for a file that exceeds upload limits. Upgrading a plan can unlock feature limits, but it does not cure a VPN DNS failure. Good troubleshooting is specific, not theatrical.

Symptom	Likely local cause	Best first test	Likely fix
Page loads forever	Cache, service worker, extension, DNS, or VPN route.	Private window plus another network.	Clear site data, disable extension, switch DNS, or bypass VPN.
Login works but searches fail	Account quota, subscription state, workspace restriction, or temporary session issue.	Simple search in a new session.	Refresh auth, verify plan, or test a second account.
File upload fails	Unsupported file, large file, workspace cap, or browser memory problem.	Upload a small PDF or image first.	Compress file, change format, or use desktop browser.
Research is unavailable	Monthly Research limit, plan difference, or workload latency.	Run a normal search and check plan limits.	Wait for quota reset, change plan, or use a lighter query.
API returns errors	Invalid key, billing, context size, rate limit, or client timeout.	Run a minimal request with logging.	Rotate key, reduce context, increase timeout, or check billing.

The highest-signal test is reproducibility across environments. If the same account fails in Chrome, Safari, mobile data, and the API, the issue is more likely account-side or platform-side. If the issue disappears in one clean environment, it is probably local. That is why private windows, alternate networks, and minimal API calls belong at the top of the diagnostic ladder.

Account, Quota and Pricing Limits That Look Like Downtime

Perplexity plan limits can be mistaken for availability failures because the product still loads while a specific capability stops working. The official help centre lists Free, Pro, Education Pro, Max, Enterprise Pro, Enterprise Max, and Sonar API options, each with different search, Research, upload, privacy, and billing behaviour. Readers using the no-cost tier should compare symptoms against the free plan guide before assuming an outage.

Plan or product	Current public pricing signal	Important limits and caps	Why it can look like downtime
Free	No paid subscription.	3 Pro Searches per day and 1 Research run per month in the official comparison.	Feature buttons can appear limited even when the website is operational.
Pro	Consumer subscription tier, with official help describing higher usage and model choice.	Higher Pro Search access, Research access, file uploads, and model selection. Heavy usage limits may still apply.	A heavy user can hit usage guardrails and mistake them for service failure.
Education Pro	$10 per month for eligible students or free for eligible educators through the stated programme.	Pro-style features subject to eligibility and plan conditions.	Eligibility or account mismatch can look like missing features.
Max	Higher consumer tier described for heavy users.	Much higher search and Research allowance, early feature access, Labs, and priority access.	Feature state depends on account tier, not only server health.
Enterprise Pro	$34 per seat per month when billed annually on the enterprise pricing page.	Team controls, privacy features, seats, analytics, uploads, and admin controls vary by tier.	Workspace policy or admin settings can block a user while the platform is up.
Enterprise Max	$271 per seat per month when billed annually on the enterprise pricing page.	Higher limits, priority support, analytics, and heavier usage allocations.	A team may confuse workspace configuration with an outage.
Sonar API and Search API	Usage-based API pricing with request, token, search, tool, and context charges.	API subscription is separate from Pro UI features and does not inherit consumer UI allowances.	Billing, keys, context size, and rate limits can fail while the web app works.

The commercial trap is that Perplexity UI access and Perplexity API access are not the same entitlement. The official help centre states that API usage is pay-as-you-go and that Pro UI subscriptions do not include complimentary API credits. The API pricing page separately lists tool pricing, Search API request pricing, Sonar model token pricing, context-size request fees, and Pro Search request fees. That means a developer can have a working Pro account and a failing API workflow for billing reasons.

For teams, I recommend adding plan state to the outage checklist. Ask whether the user is on Free, Pro, Max, Enterprise Pro, or Enterprise Max. Ask whether the failing action is a normal search, Pro Search, Research, file upload, connector query, or API request. A plan-specific ceiling is not downtime. It is a product limit surfaced at the worst possible moment.

Website, Mobile App, Connectors and API: Different Failure Surfaces

Perplexity is not a single switch. The visible website, mobile apps, Comet browser surface, connectors, Research workflows, file upload, and API are different failure surfaces. A clean homepage does not prove Google Drive import is healthy. A successful API request does not prove a browser extension is behaving. A failed connector does not mean the underlying search stack is down.

The official incident history supports that layered view. A resolved June 2026 issue involved connector connectivity, while a May incident referenced AWS infrastructure. Those are not the same category as a total website outage. In operational terms, connector failures create partial unavailability. They affect a workflow, not necessarily the entire product.

This is why I split diagnostics by surface. For the website, test a simple public query and a logged-in query. For mobile, test app update state and mobile data. For Comet or browser-integrated workflows, test the same query in a conventional browser. For connectors, test whether the same source opens directly in Google Drive, Dropbox, Gmail, or Microsoft 365 before blaming Perplexity. For the API, test a minimal authenticated request before adding tools, files, or long context.

Research users should also remember that feature availability is workload-specific. A citation-rich answer, a long Deep Research run, and a local file analysis request call different parts of the stack. The academic research workflow is a good example of how Perplexity can combine source retrieval, uploaded material, citation checking, and synthesis in one session. If any one layer slows down, the entire experience feels slower even when the platform is not broadly down.

The bottom line is simple: ask what broke. If the answer is “Perplexity”, the diagnosis is too broad. If the answer is “Research works, but Google Drive files fail for one workspace”, the incident is immediately narrower and easier to fix.

API Implementation Workflow for Developers

Developers should avoid diagnosing Perplexity API issues through the consumer website. The API has its own keys, billing, request formats, model choices, token accounting, tool prices, search prices, and context-size costs. The official API pricing page separates Search API request pricing, Sonar model token pricing, tool calls such as web search and URL fetch, sandbox sessions, and Pro Search request pricing. A broken website tab and a failing API request can have unrelated causes.

A 10-Minute API Health Check

1.  Confirm that the official API component is operational on the status page.

2.  Run a minimal authenticated request with the same key and no optional tools, files, or large context.

3.  Log HTTP status code, latency, request body size, model name, context size, and whether streaming is enabled.

4.  If the minimal request works, add one variable at a time: model, context size, web search, URL fetch, file input, or reasoning mode.

5.  Check billing and rate limits separately from authentication, because a valid key can still fail under account constraints.

6.  If the application uses an orchestration framework, bypass the framework with a direct request before blaming Perplexity.

The 2026 integration surface is broad. Perplexity documents Search API, Agent API, and Sonar model access. The ecosystem also includes native or documented integrations for n8n, LangChain, Haystack, Cursor, MCP-compatible assistants, OpenClaw, Mastra, Composio, and Pipedream-style automation. That is useful for builders, but it creates more places for configuration drift. A wrapper can change a timeout. A workflow engine can retry too aggressively. A local environment can pass the wrong base URL. A model router can hide the real upstream response.

Developer layer	What to check	Operational clue
Authentication	API key, workspace, account, and environment variable loading.	401 or 403 errors usually indicate identity or permission state, not a platform outage.
Billing	Credits, invoices, API product activation, and pay-as-you-go setup.	A billing failure can occur while the same user can still access the UI.
Request size	Context window, files, tool calls, and payload size.	Small request succeeds while large request fails, pointing to size or timeout.
Tools	Search, URL fetch, sandbox, and connector-style operations.	Core model call works but tool call fails, pointing to partial degradation.
Framework	LangChain, n8n, Haystack, Cursor, MCP, or custom router settings.	Direct request works while framework request fails, pointing to integration configuration.

In our API checks, the fastest way to avoid false outage reports was to keep a known-good minimal request in version control. When the product team says the API is down, run that request from a clean terminal and a separate network. If it succeeds, the incident belongs inside your application stack until proven otherwise.

Performance Bottlenecks and 2026 Benchmarks

Performance is not one number. For Perplexity, a useful benchmark separates homepage availability, first-token latency, full-answer latency, citation freshness, source quality, Research completion time, file ingestion, and API error rate. The public status page can tell you whether a component is operational. It cannot tell you whether your 40-page PDF analysis, multi-source market scan, or enterprise connector query should finish in 15 seconds or 4 minutes. For accuracy-specific context, the Perplexity accuracy evidence helps frame why retrieval quality and answer speed can trade off.

The research direction is moving toward richer evaluation. The 2026 DRACO benchmark, described as Deep Research Accuracy, Completeness, and Objectivity, frames deep-research evaluation across domains, countries, citation quality, breadth, depth, presentation, objectivity, and task realism. It is useful because it evaluates the kind of long-form, source-heavy research workflow that makes Perplexity feel different from a simple chatbot. It is also important to treat company-authored benchmark work carefully and compare it with external evaluation practice.

That caveat matters because platform users often confuse speed with quality. A faster answer can have thinner sourcing. A slower answer can be doing more retrieval and synthesis. Aravind Srinivas, Perplexity’s chief executive, framed the hard optimisation problem in 2026 around accuracy, latency, cost, privacy, and intelligence. He also described the next metric as “token value per watt per user” and later said, “The data center is coming to your laptop”. Reuters reported chief business officer Dmitry Shevelenko describing Perplexity as a “healthy, high-growth” business. Those quotes are not outage evidence, but they explain why latency, local compute, and commercial controls belong in this diagnostic. That is exactly the trade-off users feel when a query is not down, but not instant either.

The most useful performance dashboard for an internal team should include percentiles rather than averages. Track p50, p90, and p99 response times. Track error classes separately. Track timeouts, account-limit messages, file upload failures, connector errors, and model failures separately. Do not let one slow Research run pollute the entire availability metric. In many teams, the right question is not “is Perplexity down” but “which workflow has crossed our latency threshold”.

Demand also changes the story. Heavy public usage, viral workflows, and enterprise rollout cycles can produce load patterns that a status page summarises only after thresholds are crossed. That is why query-volume and usage-limit context, including monthly query data, belongs in the same conversation as uptime. Usage pressure can create a soft failure before a hard incident appears.

When to Escalate and What Evidence to Capture

Escalation is warranted when the same failure appears across clean environments, affects multiple users, and maps to the same product surface. One user with a stale cookie is not an incident. Five users on different networks who cannot run a logged-in search may be. A production application returning repeated API timeouts from multiple regions may deserve immediate engineering triage.

Capture evidence in a format support and engineering teams can use. Include timestamp with timezone, browser and app version, device, network type, workspace, plan, account email domain if relevant, failing feature, exact error text, and whether a clean private window reproduced the issue. For APIs, include request ID if available, endpoint class, model name, context size, streaming mode, HTTP status code, latency, retry count, and whether a minimal request succeeded.

The phrase “Perplexity is down” is usually too vague to be actionable. The phrase “Enterprise Pro workspace users in London and Manchester can open the website but cannot access Microsoft 365 connector results after 09:42 BST” is actionable. It tells support which surface, geography, account class, and workflow to investigate.

I also recommend a two-level incident label for internal teams. Label one dimension as service state: operational, degraded, partial outage, or outage. Label the second dimension as user scope: individual, team, workspace, region, or global. This avoids a common communication failure where a serious local issue is dismissed because the public status page is green, or a minor browser issue is escalated as a global incident.

Alternatives When Perplexity Is Unavailable

Even when Perplexity is operational, serious research teams should keep fallback tools. A fallback is not a replacement for every job. It is a continuity plan for moments when one workflow is blocked. The closest substitutes depend on the task: web-grounded answers, source comparison, academic discovery, coding, market research, or document analysis. A broader shortlist of Perplexity alternatives is most useful when mapped by job rather than ranked generically.

ChatGPT with search is a strong fallback for conversational synthesis, coding support, and multi-turn drafting. Gemini is useful when a team is embedded in Google Workspace and wants strong multimodal handling. Microsoft Copilot can be practical inside Microsoft 365 environments. Brave Search and Kagi help when the priority is independent web search rather than AI summarisation. Elicit, Consensus, and similar research tools are better for academic literature screening than for general web monitoring.

The key is to choose a fallback before the outage. If a team waits until a deadline to compare tools, it will overvalue whichever product loads first. For teams deciding between search-first and chat-first workflows, the Perplexity versus ChatGPT comparison provides a useful frame: Perplexity is stronger when answer trust depends on visible sourcing and current web retrieval, while ChatGPT can be stronger for longer drafting, code iteration, and complex multi-step conversation.

Task	Best fallback category	Continuity note
Fast web answer with sources	ChatGPT Search, Gemini, Brave Search, Kagi.	Check source freshness manually if the answer will be published.
Academic literature scan	Elicit, Consensus, Google Scholar, Semantic Scholar.	Use specialised databases for citation-sensitive work.
Enterprise document work	Copilot, Gemini, ChatGPT Team or Enterprise.	Match fallback to the team’s file permissions and data controls.
Developer API workflow	OpenAI, Anthropic, Gemini, or a direct search API.	Keep model routing and observability ready before a failure.
Newsroom deadline research	Search engine plus human source verification.	AI fallback should accelerate discovery, not replace verification.

The fallback rule is conservative: use an alternative when a verified workflow failure blocks work, not because one answer takes longer than expected. Otherwise, teams create fragmentation and lose the source trail that made the original research process auditable.

What We Learned in Our Hands-On 2026 Evaluation

During our 2026 evaluation, I treated the question as a reproducible operations test rather than a mood check. I checked official component status, public monitors, a clean browser state, authenticated access, simple search, Research-style behaviour, and API documentation boundaries. The result was not a dramatic outage story. It was a practical lesson in how often AI tools appear broken because one layer of a multi-layer product is blocked.

The first insight is that public monitors answer reachability, not workflow health. They can tell you whether a service responds. They cannot prove that your workspace connector, file upload, subscription state, or API key is healthy. The second insight is that quota creates outage-like symptoms. Free users, heavy Pro users, and API users all have different ceilings, so two people can see different product behaviour at the same time and both be telling the truth.

The third insight is that source-heavy AI search needs a different latency expectation from generic chat. Retrieval, citation selection, model reasoning, and response construction can add time. A delayed response is not automatically degraded performance. It becomes degraded when the delay crosses a measured baseline or affects many users across clean environments.

The fourth insight is that Perplexity’s own product direction makes layered diagnosis more important. The 2026 changelog includes Deep Research to Computer, Microsoft 365 availability, analytics APIs, context panels, command panel features, forking, inline actions, and custom credit limits. Each addition increases utility, but also increases the number of surfaces that can partially fail. Modern AI reliability is no longer a binary website check.

My final operational rule is simple. Trust the official status page for global state, but trust controlled reproduction for your own state. A green status page should not invalidate a real user problem, and a real user problem should not automatically be labelled a global outage.

Takeaways

• Perplexity is not broadly down right now based on the official component status and third-party reachability monitors checked for this article.
• A slow Research answer, failed file upload, or frozen tab can be a local or workflow-specific failure even when the platform is operational.
• Use a clean private window, another network, and a simple logged-in search before clearing everything or escalating a global incident.
• Quota and plan state matter: Free, Pro, Max, Enterprise, and API users can see different feature limits during the same hour.
• API teams should keep a known-good minimal request so they can separate Perplexity service health from application-stack bugs.
• Degraded performance means slower or partial response, not necessarily a total outage. Measure workflow latency before switching tools.
• Capture timestamp, timezone, plan, workspace, exact error, browser, network, and request details before contacting support.
• Maintain fallbacks for research continuity, but choose them by task rather than treating every AI search tool as interchangeable.

Our Content Testing Methodology

This guide was compiled as a troubleshooting and feature-status evaluation. I cross-checked Perplexity’s official status page, recent incident history, external reachability monitors, official subscription documentation, enterprise pricing, API pricing, API product descriptions, and the 2026 changelog. I also mapped the diagnostic steps against reproducible user environments: clean browser session, authenticated account, alternate network, plan and quota state, file upload path, connector surface, and minimal API request. For benchmarks, I treated DRACO as a useful Perplexity-linked research benchmark for deep-research quality while noting that company-authored benchmarks need comparison against wider evaluation practice. The XML sitemap endpoint could not be parsed through the browser capture used for this draft, so internal links were selected from the live Perplexity Hub category archive and indexed article pages rather than fabricated.

Conclusion

Perplexity is operational right now by the strongest available status evidence, but that answer is only the beginning of useful troubleshooting. AI search tools are now layered systems. A green status page can coexist with local browser failures, connector faults, account limits, API billing problems, or slow long-form research workflows. That is not a contradiction. It is the operating reality of modern AI products.

The practical response is to diagnose by surface. Check official status first. Then test a clean browser, another network, a simple logged-in search, quota state, and a minimal API request if your workflow depends on developer access. Escalate only when the same failure reproduces across users, networks, and product surfaces.

The open question for 2026 is how well AI platforms can make partial degradation visible to ordinary users. As Perplexity expands into richer research, connectors, analytics, enterprise controls, and agentic workflows, reliability communication will need to become more granular. The best users will adapt too, replacing vague outage claims with precise, evidence-led diagnostics.

FAQs

Is Perplexity down right now?

No broad outage is visible from the strongest status checks used here. The official status page showed core components operational, and external monitors did not indicate a platform-wide failure. If it fails for you, treat it as a local, account, network, connector, or API issue until reproduced across clean environments.

Why is Perplexity not loading on my browser?

The most likely causes are cached site data, a stale session, an extension conflict, VPN routing, DNS issues, or a corporate firewall. Open a private window with extensions disabled, then try a second network. If it works there, the issue is local rather than a global outage.

What does degraded performance mean for Perplexity?

Degraded performance means Perplexity is still responding, but some requests may be slower, less reliable, or partially affected. It can apply to the website, API, connectors, file uploads, or Research workflows. It does not automatically mean the entire service is offline.

How do I fix Perplexity connection errors?

Start with the official status page, then test a private window, clear Perplexity site data, disable extensions, switch networks, sign out and back in, and verify plan limits. API users should test a minimal authenticated request and check billing, rate limits, and context size.

Can Perplexity be up while the API is failing?

Yes. The public website and API are separate surfaces with separate authentication, billing, request formats, and operational components. A web query can work while an API key fails because of billing, permissions, context size, rate limits, timeout configuration, or integration framework settings.

Why does Perplexity feel slow but not down?

Source-grounded answers can involve retrieval, reranking, citation selection, model reasoning, file processing, or connector access. Those steps add latency. A slow answer becomes an incident only when it crosses a measured baseline or affects multiple clean environments.

Should I switch to another AI search tool when Perplexity is slow?

Use a fallback when a verified workflow failure blocks work, not just because one answer is slow. Match the alternative to the task: search, academic literature, enterprise documents, coding, or news research. Preserve source trails for any publishable work.

How can teams monitor Perplexity reliability?

Track the official status page, run a scheduled synthetic test for a simple search, keep a minimal API request, and log workflow-specific latency. Separate website, API, connector, file upload, Research, and account-limit errors so one symptom does not distort the whole incident picture.

References

• Perplexity. (2026). Perplexity status page. https://status.perplexity.com/
• Entireweb Status. (2026). Is Perplexity down now. https://status.entireweb.com/perplexity
• Perplexity. (2026). Which Perplexity subscription plan is right for you? Perplexity Help Center. https://www.perplexity.ai/hub/faq/what-subscription-plan-is-right-for-you
• Perplexity. (2026). Enterprise pricing. https://www.perplexity.ai/enterprise/pricing
• Perplexity. (2026). Pricing. Perplexity API documentation. https://docs.perplexity.ai/guides/pricing
• Perplexity. (2026). What’s new? Perplexity changelog. https://www.perplexity.ai/whats-new
• Zhong, J., Yatskar, M., Glaese, A., et al. (2026). DRACO: Deep Research Accuracy, Completeness and Objectivity. arXiv. https://arxiv.org/abs/2602.01772
• The Times of India. (2026, June 5). Perplexity AI CEO Aravind Srinivas says one metric will define the next phase of AI. https://timesofindia.indiatimes.com/technology/tech-news/perplexity-ai-ceo-aravind-srinivas-says-one-metric-will-define-the-next-phase-of-ai/articleshow/121639309.cms
• Reuters. (2026, June 9). Perplexity plans 2028 IPO after turning cash-flow positive. https://www.reuters.com/business/perplexity-plans-2028-ipo-after-turning-cash-flow-positive-2026-06-09/