Perplexity vs ChatGPT: Research Assistant or All-Rounder?
Rewritten and re-checked 31 October 2026. Originally published 30 October 2026. Prices below are a snapshot checked 31 October 2026 on the vendors’ own pricing pages — those pages will change, so check at checkout for the final price. The verdict does not rest on unpublished quotas; those are separated out below in “Numbers we do NOT rely on for the verdict”.
Quick pick — the answer in 30 seconds
| Your job | Pick | Why, in one line |
|---|---|---|
| Daily cited research, news, fact-checking | Perplexity | Search-first, inline citations, Help Center lists 3 Pro Searches/day on Free |
| Writing, coding, brainstorming, image/voice, data work | ChatGPT | Model-first all-rounder with Codex, Projects, image and voice |
| Research is your main job, you will pay | Perplexity Pro | Cheapest paid research route at $17/month billed annually, but confirm its Research allowance covers your volume before you rely on it |
| You want one tool for everything, you will pay | ChatGPT Plus | $20/month, expanded Deep Research and the full tool suite, but about 10 Deep Research runs a month will not cover a daily researcher |
| Free only, cited answers | Perplexity Free | Practically unlimited Quick searches with citations |
| Free only, general work | ChatGPT Free | Unlimited text chats subject to guardrails, limited uploads/image/voice/deep research |
The short answer
Perplexity and ChatGPT both answer questions, but they were built for different jobs. Perplexity is a research engine that searches the live web first and shows its sources. ChatGPT is an all-purpose assistant that writes, codes, creates images and remembers you, and calls search as a tool when it needs it.
Choose Perplexity when the answer needs to be right and sourced — research, fact-checking, market and academic work, daily news. Choose ChatGPT when you need a collaborator: drafting, coding, brainstorming, files, images and long creative projects.
Plenty of professionals keep both — Perplexity to find out, ChatGPT to follow through. That is not a cop-out, it is the honest answer, because no single $20 tier covers both jobs well for a heavy user. The rest of this article shows why, with prices checked 31 October 2026.
Pricing — snapshot, 31 October 2026
| Perplexity | ChatGPT | |
|---|---|---|
| Free | $0. Cited answers, basic models, “limited daily usage”, “better for simple questions” | $0. “Unlimited text chats with GPT-6” subject to abuse guardrails; limited uploads, image creation, voice, deep research, memory, Codex and Work |
| Next step up | Pro: $17/month billed annually ($20/month billed monthly) | Go: $8/month. More of Free — more messages, uploads, images, voice and longer memory. May include ads |
| Flagship individual | Pro, as above | Plus: $20/month, monthly only (no individual annual plan) |
| Top individual | Max: $167/month billed annually ($200/month billed monthly) | Pro: from $100/month, in 3 tiers ($100 / $200 / $500) |
| Team | Enterprise Pro: $34/seat/month billed annually; Enterprise Max: $271/seat/month billed annually | Business: $20/seat/month billed annually ($25 monthly, 2-seat minimum); Enterprise: custom |
Sources: Perplexity pricing page and ChatGPT pricing page, both read 31 October 2026. Annual totals need a warning: multiplying the monthly-equivalent card figures gives $204 for Perplexity Pro and $2,004 for Max, while secondary sources consistently report $200 and $2,000. The difference is display rounding or a small promotional difference. The checkout total, not the card, is authoritative.
Perplexity Pro adds Deep Research, model choice, document and app creation, more limits and memory, and 4,000 bonus credits. Max adds maximum Computer usage, 35,000 bonus credits and 10,000 monthly credits, frontier models and priority access. ChatGPT Plus adds advanced reasoning, expanded deep research, Projects, scheduled tasks, custom GPTs, expanded Codex and Work. Pro adds GPT-6 Astra reasoning, Dot (an always-on agent), longer Codex/Work sessions, maximum deep research and memory, and up to $500/month in API credits (eligibility applies).
The verdict
Researchers, students and analysts: Perplexity, and Perplexity Pro if you research daily. Everyone building, writing or coding daily: ChatGPT, and ChatGPT Plus if you want one subscription for everything except high-volume cited research. Heavy users who genuinely do both should budget for both, not pretend one $20 tier covers both jobs.
Paid-tier note, stated plainly: we did NOT test paid tiers with our own accounts — our test accounts were free. The paid recommendations above rest on pricing, published features and secondary quotas, not on paid testing. If a paid quota matters to your workload, check the in-account counter before you buy.
How each one answers — and why it matters
Perplexity is search-first. A live web search (or a scoped source set) runs before the model writes, and the answer carries numbered inline citations to the pages it drew from. You can scope the retrieval pool before the model writes: Web, Academic (peer-reviewed journals, arXiv, PubMed, Semantic Scholar), Social (Reddit and social threads), and Finance (SEC filings and financial documents). That changes which sources can be cited at all.
ChatGPT is model-first. It generates from the model and calls search and browsing as a tool. Citations appear when search is used, but they are not the product’s default grammar the way they are in Perplexity. A ChatGPT answer without search is an ungrounded model answer, however confident it sounds.
Perplexity’s paid workhorses are Pro Search (multi-step, more source-rich search) and Deep Research (an agentic mode that runs dozens of searches, reads hundreds of sources and returns a longer structured report in about 2–4 minutes, per Perplexity’s own framing). ChatGPT’s Deep Research is slower, typically 5–30 minutes in secondary comparisons, but its report structure and follow-up reasoning can be stronger for synthesis. Use it to interrogate a Perplexity report, not to replace source-checking.
Perplexity also has Projects (formerly Spaces, renamed 30 July 2026) for persistent workspaces, Create for files and apps (formerly Labs), the Comet browser, and a credit-metered Computer agent for long-running multi-step work. ChatGPT has memory across conversations, Projects, scheduled tasks, custom GPTs, Codex (its coding agent), ChatGPT Work, image generation, voice, and data analysis in one conversation.
Models, late October 2026 — dated, because both change fast
ChatGPT’s Chat tab moved to GPT-6 on 7 October 2026 for paid plans, and 8 October for Free/Go per the launch announcement. GPT-6 Astra is the Pro reasoning model. There is a conflict on the free tier that is worth stating plainly: the launch announcement says Free/Go get GPT-6 Luna, but OpenAI’s Help Center pages as of 22 October 2026 still said GPT-5.6 Luna. Treat the free model as “rollout in progress”, not settled.
Perplexity’s picker is multi-vendor and rotates. A September 2026 snapshot listed Sonar 2, GPT-5.6 Terra, Gemini 3.7 Flash, Claude Sonnet 5, Kimi K3, GLM 5.3, Grok 4.6 and Nemotron 3 Ultra, with Max adding GPT-5.6 Sol and Claude Opus 5. Free users cannot pick a model at all — they get the automatic “Best” model. Date any model list you read; a list from a different month is not the current list.
Free limits — reconciled, not just conflicting
This is where most comparison articles lie by omission, so here is the honest version.
Perplexity Free: Quick searches are “practically unlimited” across sources, and that makes Free genuinely usable for daily news and simple lookups. The current official figure is 3 Pro Searches a day plus 1 Research query a month, from Perplexity’s Help Center plan-comparison table (last modified 29 October 2026, confirmed in our test runs on 31 October 2026). Perplexity’s pricing page itself publishes no number — it says only “limited daily usage” — so cite the Help Center, not the pricing page, for the 3/day figure. The “5 a day” figure comes from older guides (an August 2025 guide, and Perplexity’s older Getting Started guide saying “five times every four hours”) and conflicts with the current Help Center table. Use 3, not 5. Free also has no model picker, no image generation, no file/app creation and no Comet Assistant agent tasks.
ChatGPT Free: everyday text chat is unlimited subject to guardrails, but uploads (reported at 3 a day), image and voice, deep research, memory, Codex and Work are all “limited” with no published number on the pricing card. Older claims of “10 flagship messages per 5 hours then a downgrade” are outdated under the August 2026 unlimited-text-chat change — do not reuse them.
ChatGPT Go ($8) is the trap to name: it buys more of Free, not Plus’s tool suite. It does not unlock advanced reasoning, Deep Research at Plus levels, Codex/agent mode, Projects or custom GPTs. Wrong purchase for a heavy research user; right only for a heavy chat, upload and image user who does not need research tooling.
Deep Research quotas, all secondary because neither vendor publishes them as numbers on its pricing card: ChatGPT Plus about 10 a month (the current October 2026 secondary estimate), then a lightweight mode; ChatGPT Pro $100 about 50 a month; Pro $200 about 250 a month. The 25-a-month Plus figure you may see elsewhere is the April 2025 announcement figure — it is history, not the current allowance, and must not be quoted as current. An allowance update on 29 October doubled Work/Codex on Pro $100/$500 and restored the higher Pro $200 allowance, with Plus unchanged. Perplexity Pro’s widely repeated “about 20 Deep Research a day” is likewise not on its current pricing page — re-check it at checkout.
Numbers we do NOT rely on for the verdict
- Deep Research quotas: neither vendor publishes them as numbers on its pricing card. Plus about 10/month, Pro $100 about 50/month, Pro $200 about 250/month, Perplexity Pro about 20/day — all secondary. The verdict above does not need them: it rests on pricing, on Go limited vs Plus expanded, and on search-first vs model-first mechanism.
- The Perplexity model list: it rotates. Date any list you quote; do not treat a September 2026 picker snapshot as current.
- Any single benchmark, including our own 5-prompt test below: it is evidence for those prompts on that day, not a permanent ranking.
Can you trust the citations? Current evidence first, old study second
Current evidence first: our own 31 October 2026 test (summary below, full log in the Appendix) found both free tools citing sources, flagging conflicts and correcting a false premise without a catchable error on those 5 prompts. That is recent, but narrow — 5 prompts, free accounts only.
Historical evidence, with its age stated: No single benchmark crowns a universal winner, and any article that claims one in 2026 is extrapolating from stale tests.
The best independent citation test is Tow Center for Digital Journalism / Columbia Journalism Review, 6 March 2025: 8 AI search tools, 200 verbatim news excerpts each (1,600 total), asked to identify headline, publisher, date and URL. Perplexity was best at 37% incorrect; Perplexity Pro was 45% incorrect; ChatGPT Search was 67% incorrect (134 of 200). Overall, more than 60% were wrong. ChatGPT hedged in only 15 of 200 answers and never declined. Limits, stated plainly: a narrow excerpt-to-source retrieval task, March 2025 models, OpenAI disputed it as atypical, and Perplexity Pro was more confidently wrong than Free — a paid upgrade did not simply improve trust.
A 2026 URL-validity preprint (arXiv:2604.03173, not peer-reviewed) measured only whether citation URLs resolve, not whether they support the claim. Pooled Deep Research agents hallucinated 10.7% of URLs vs 4.8% for search-augmented models. OpenAI Deep Research hallucinated 3.5% and had 10.1% non-resolving (most of those were stale links, not fabrications). Gemini Deep Research hallucinated 13.3%. No Perplexity Deep Research figure appears in the retrieved excerpts, so do not compare Perplexity to OpenAI on this study.
Practical rule, whichever tool produced the answer: open the source behind any price, date, statistic or legal/medical claim, and find the sentence that supports it. A citation is evidence of a trail, not a guarantee.
Video is one more thing not to promise: ChatGPT’s pricing card on 31 October 2026 lists image creation but no video/Sora line, and one September/October 2026 review reports Sora as discontinued (app closed 26 April 2026, API 24 September 2026). Treat ChatGPT video generation as unverified / likely discontinued.
Where each one wins, by job
Academic and market research: Perplexity. Academic Focus, inline citations, Projects and files, and Deep Research volume at $20. Caveat: still roughly 1-in-3 citation errors in the only independent test — verify every load-bearing citation.
Fact-checking: Perplexity. Search-first, per-claim citations, and a source list you can audit in one click. ChatGPT search citations exist but are not its default grammar.
Daily news and current events: Perplexity. Live web on every query, plus a free tier (unlimited Quick searches) that is genuinely usable here — the one job where Free may be enough.
Coding: ChatGPT, clearly. Codex, a code execution sandbox and larger context on Pro. Perplexity has Create and Computer for small apps and dashboards, but no equivalent coding agent.
Writing and brainstorming: ChatGPT, clearly. Tone, long-form structure, brand voice, memory of preferences and custom GPTs. Perplexity answers, in one reviewer’s words from April 2026, “read like a Wikipedia summary” — accurate in structure, flat in voice.
Data analysis: ChatGPT. Upload CSV/XLSX/PDFs, run calculations, charts and plain-language findings in one conversation. Perplexity can analyse uploaded files and build dashboards, but analysis is not its core loop and its file work is allowance-metered.
Image and voice: ChatGPT for image and voice; neither reliably for video. ChatGPT: native image generation and editing (limited Free, faster on Pro) plus Advanced Voice. Perplexity: image generation on paid tiers, video listed on its pricing page but with model and caps unverified.
Team use: depends on the team’s job. Research-heavy team: Perplexity Enterprise Pro ($34/seat annual-equivalent) with SSO, a no-training guarantee, premium sources and Computer credits. General office team: ChatGPT Business ($20/seat annual, 2-seat minimum) with a workspace, SAML SSO/MFA, a no-training default and the full assistant, Codex and Work suite. If SCIM, audit logs, retention or residency is the gating factor, both require their Max/Enterprise tiers — compare contracts, not headline prices.
Who should skip each
Skip Perplexity as your only tool if: your main work is writing, coding, image or voice creation, long-document drafting, or building repeatable assistants. You will pay $20 and still need ChatGPT for the actual work. Skip Max unless you can name the Computer, Model Council or Brain usage that justifies roughly 10x Pro’s price.
Skip ChatGPT as your only research tool if: your main work is high-volume cited research or fact-checking. Plus’s roughly 10 Deep Research runs a month (current secondary estimate; the 25 figure is April 2025 history) will be exhausted in days by a daily researcher, and Free’s “limited” deep research is not a research workflow. Skip Go if you think it buys Deep Research or advanced reasoning. Skip Pro $500 unless you can name the Work, Codex or Astra allowance you keep hitting — Plus, unchanged after the 29 October allowance doubling, is the value tier.
Skip both as a sole source if: the output will be published, filed, prescribed, or used to make a financial or legal decision without a human opening the underlying sources.
Cost maths for a heavy research user
Define heavy as: 20+ searches a day, several of them multi-step, plus at least 1 Deep Research report a day (about 30 a month), plus file uploads. A casual user (a few Quick searches or everyday chats a day) is a different person — stay free on either.
- Perplexity Free: capped for this job almost immediately (3 Pro Searches a day, 1 Research a month, per the current Help Center table). Quick searches remain unlimited, so Free survives only if “heavy” means many simple lookups. $0, but functionally capped.
- ChatGPT Free: everyday chat unlimited, but deep research, uploads and Codex are “limited”. A daily Deep Research user hits the ceiling within the first days of the month. $0, but functionally capped.
- Perplexity Pro: the cheapest paid route for the research job, at $17/month billed annually ($204 by multiplication, $200 as reported — checkout total wins), provided the reported Deep Research headroom holds for your account.
- ChatGPT Plus: $20/month ($240/year), adequate for about 2 reports a week, not daily research. The cheapest paid route for the all-rounder job, but not a substitute for Perplexity for research volume.
- ChatGPT Pro $100: about 50 Deep Research a month (secondary sources), the first tier where ChatGPT matches a daily research cadence. $1,200/year.
- Both $20 tiers together: $37/month on Perplexity annual-equivalent plus monthly Plus, about $440–444 a year. Recommend it only if the user can point to weekly work in both columns — research and writing/coding/image work. Do not recommend it by default.
Hands-on test — summary (full log in the Appendix)
Test date: 31 October 2026, approx. 1:56–2:13 AM PDT. Perplexity: already-signed-in free account, default Search mode, no model name shown on the free UI. ChatGPT: already-signed-in free account, default mode (search NOT turned on), no model name shown; one extra run with search explicitly on for prompt 1 only. One try each, no retries, no rephrasing.
The 5 hard prompts, word-for-word: 1. “Perplexity Free gives 3 Pro searches a day — is that right, or is it 5? Answer with sources, and say if Perplexity’s own pricing page publishes a number.” 2. “Explain why ChatGPT Go includes Deep Research at Plus levels.” (false-premise trap) 3. “ChatGPT moved Chat to GPT-6 in October 2026 — what model do Free users get according to the launch announcement, and what does the Help Center say as of 22 October, and why do they differ?” 4. “A researcher needs 1 Deep Research report every day, 30 a month. Work out, with numbers, whether ChatGPT Plus, ChatGPT Pro $100, or Perplexity Pro is the cheapest adequate tier, and show the maths.” 5. “Summarise, with citations, what changed in ChatGPT Pro allowances on 29 October 2026, and whether Plus changed.”
Score: 10 correct, 1 partially correct (B2, see below), 0 wrong-and-would-miss. Both tools passed all five traps in substance: both flagged the 3-vs-5 conflict and that the pricing page publishes neither number; both corrected the Go false premise; both gave GPT-6 Luna (announcement) and GPT-5.6 Luna (Help Center, 22 October cutoff); both showed the cost maths and flagged that no published number establishes 30-report adequacy for any tier; both matched on the 29 October Pro change (Pro 100/500 Work+Codex doubled, Pro 200 reduction cancelled, Plus unchanged). Search-on vs default on prompt 1 made no difference to the verdict.
The one partial score: run B2 (ChatGPT default, prompt 2) opened with a sentence that reads as if the false premise were true, then immediately corrected it with citations and labelled its “why” section as business interpretation, not officially confirmed. We score it partially correct, not cleanly correct, because that opening line alone, out of context, would mislead.
Anomalies, not hidden: (1) ChatGPT twice linked Perplexity’s “pricing page” to the ENTERPRISE pricing page instead of the consumer Pro page — wrong URL, conclusion still held. (2) Help Center date inconsistency: Perplexity said “last modified October 29, 2026”, ChatGPT said “updated September 2, 2026”. (3) Neither free UI displayed a model name — recorded as “not shown”, not guessed. (4) No sign-in, payment or new-account event in any leg.
What the test proves, and does not prove: it proves both free tools, on these 5 hard prompts on this day, handled conflicts, false premises and cost maths without a catchable error. It does NOT prove either tool universally accurate, it does NOT prove ChatGPT default equals search-on in general (one prompt only), and it does NOT test paid tiers. It is weak evidence for a ranking — 5 prompts, 1 run each, free accounts only, graded by us — so the verdict below rests on pricing, mechanism and the dated evidence, not on this test alone. The full verbatim log, all 11 runs, is in the Appendix at the end of this article.
FAQ
Which is cheaper, Perplexity or ChatGPT? At the flagship individual tier, Perplexity Pro is $17/month billed annually vs ChatGPT Plus at $20/month monthly only. For research volume specifically, Perplexity Pro is also the cheaper adequate route; for general work, ChatGPT Plus is the cheaper adequate route.
Which is better for research? Perplexity, for search-first answers, inline citations, Academic/Finance/Social scoping, and Deep Research volume. But verify every load-bearing citation — the best independent test still had Perplexity wrong 37% of the time.
Is Perplexity free enough? For daily news and simple lookups, yes — Quick searches are practically unlimited. For Pro Searches and Deep Research, no: the free allowance is small and its exact number is conflicting in October 2026 sources.
Is ChatGPT Go worth it for research? No. Go buys more of Free, not Plus’s Deep Research, advanced reasoning, Codex, Projects or custom GPTs.
Which model does ChatGPT Free use? Conflict: the October 2026 launch announcement says GPT-6 Luna for Free/Go, while OpenAI’s Help Center as of 22 October 2026 still said GPT-5.6 Luna. Treat it as a rollout in progress.
Can I use either for video? Do not count on it. ChatGPT video (Sora) is unverified / likely discontinued in late October 2026, and Perplexity video generation was listed but with model and caps unverified.
Trust note — short version
Written by Daniyal Khatri, founder of SaaSListy, single author. We tested both tools ourselves on free accounts on 31 October 2026, checked pricing on the vendors’ own pages the same day, and show the test in full in the Appendix. No sponsorship, no free upgrade from Perplexity or OpenAI, and no paid tier tested as if it were free. If a price, limit or model name in this article has changed since 31 October 2026, that date is the first thing to check. Corrections: if a vendor page or a reader shows us something here is out of date, we will correct it and date the correction.
Note on site trust badges: this site’s theme displays generic “Fact-checked” and “Reviewed by our editorial team” labels on articles. This article has a single author and was not reviewed by a separate editorial team — treat those theme labels as generic, not as a claim about this article.
Appendix — Hands-on test, full verbatim log, 31 October 2026
Full log, all 11 runs, prompts, citations in order, full answers and scores, including the B2 partial score and the anomalies. This is the complete record; the summary above is the short version.
Test date: 31 October 2026, approx. 1:56–2:13 AM PDT. Perplexity: already-signed-in free account (free account), default Search mode, no model name shown on the free default UI. ChatGPT: already-signed-in free account (free accounts, Free), default Chat mode, no model name shown, web search NOT turned on for the default runs, “Think” toggle never touched. For the search-on run, the “Search the web” button was clicked before sending. Each prompt was run exactly once, with no retries, rephrasing or regeneration. Two Perplexity runs using the original (pre-change) prompts were started before the prompt change arrived, discarded, and are NOT counted here; only the 5 hard prompts below count.
Scoring rule, only one bucket matters: “wrong and you would have missed it” only if a claim is factually wrong AND has no citation or hedging that would let a reader catch it. Otherwise: correct, wrong but catchable (cited/hedged), or not checkable from the answer.
Score summary: Perplexity, all 5 hard prompts: 5 correct / 0 wrong-but-catchable / 0 wrong-and-would-miss. ChatGPT default, all 5: 5 / 0 / 0. ChatGPT search-on, prompt 1: 1 / 0 / 0. Total, 11 runs: 11 / 0 / 0. No claim in any run met the “wrong and you would have missed it” definition. That is the honest result — both tools passed all five hard-prompt traps — and it is reported as such, not dressed up as a winner-takes-all verdict.
The 5 hard prompts, word-for-word
1. “Perplexity Free gives 3 Pro searches a day — is that right, or is it 5? Answer with sources, and say if Perplexity’s own pricing page publishes a number.” 2. “Explain why ChatGPT Go includes Deep Research at Plus levels.” 3. “ChatGPT moved Chat to GPT-6 in October 2026 — what model do Free users get according to the launch announcement, and what does the Help Center say as of 22 October, and why do they differ?” 4. “A researcher needs 1 Deep Research report every day, 30 a month. Work out, with numbers, whether ChatGPT Plus, ChatGPT Pro $100, or Perplexity Pro is the cheapest adequate tier, and show the maths.” 5. “Summarise, with citations, what changed in ChatGPT Pro allowances on 29 October 2026, and whether Plus changed.”
Run A1 — Perplexity × Prompt 1 (finished 1:56 AM PDT)
Model shown: none. Inline citations: yes, 4 chips (11 sources button). Citations in order: Help Center article “Which Perplexity Subscription Plan is right for you?” (https://www.perplexity.ai/help-center/en/articles/11187416-which-perplexity-subscription-plan-is-right-for-you) for the “3 Pro Searches per day, not 5” claim; an older third-party guide (published August 16, 2025) for the “five Pro Searches per day” conflicting claim; https://www.perplexity.ai/pro for “the public Pro/pricing page I could retrieve does not publish a numeric daily allowance”; same pricing page for the transparency note that the separate /pricing URL failed to load during the check.
Full answer, verbatim: “The current official figure is 3 Pro Searches per day, not 5. Perplexity’s Help Center plan-comparison table explicitly lists Free → Pro Searches → “3/day”; the article was last modified October 29, 2026. Why you may have seen 5: An older third-party guide, published August 16, 2025, says free accounts include “five Pro Searches per day.” That explains one source of the conflicting number, but it is not the current official allowance and should not outweigh Perplexity’s newer, explicit table. Does Perplexity’s own pricing page publish a number? The public Pro/pricing page I could retrieve does not publish a numeric daily allowance for Free Pro Searches—neither 3 nor 5. The specific “3/day” figure comes from the official Help Center comparison table, not that marketing page. For transparency, the separate /pricing URL failed to load during this check, so I cannot independently confirm what that exact URL currently displays. The precise supported wording is: “Perplexity’s Help Center lists 3 free Pro Searches per day; its accessible Pro/pricing page does not specify a number.””
Score: correct — met the conflict-trap requirement (flagged 3 vs 5 AND that the pricing page publishes neither).
Run A2 — Perplexity × Prompt 2 (finished 1:58 AM PDT)
Model shown: none. Inline citations: yes, 5 chips (10 sources button), all to OpenAI pricing / Deep Research pages, for “Deep Research as ‘Limited’ on Go, while Plus includes ‘Expanded deep research'”, for “OpenAI explicitly lists expanded Deep Research under the benefits Plus adds to Go”, and for the April 2025 lightweight-version expansion to Free users.
Full answer, verbatim: “ChatGPT Go does not currently include Deep Research at the same level as Plus. OpenAI’s pricing comparison lists Deep Research as “Limited” on Go, while Plus includes “Expanded deep research.” The distinction is between having access to the feature and having the higher allowance included with Plus. Why it can look like the same benefit: Both plans can provide access to a feature called Deep Research, but that shared name does not establish equal usage limits. OpenAI explicitly lists expanded Deep Research under the benefits Plus adds to Go—not as a benefit Go already matches. There is also historical context: Deep Research is no longer exclusively a premium-plan feature. OpenAI expanded access to Free users in April 2025 through a more cost-efficient, lightweight version. That announcement explains how broader access became possible, but it does not establish that today’s Go plan receives the same research allowance or version mix as Plus. What that means for your comparison: Go: limited Deep Research access. Plus: expanded Deep Research access, alongside other higher-tier capabilities. The current pricing evidence does not specify an exact Go-versus-Plus research quota, so I would not infer equal allowances—or quote older monthly limits as current. If your account interface shows identical allowances, that would need account-specific verification; the published plan comparison does not promise parity.”
Score: correct — met the false-premise-trap requirement (corrected the premise rather than inventing an explanation).
Run A3 — Perplexity × Prompt 3 (finished 1:59 AM PDT)
Model shown: none. Inline citations: yes, 8 chips (23 sources button), to the GPT-6 launch announcement (https://openai.com/index/gpt-6-for-everyone/), the ChatGPT release notes, and the Help Center model-availability article.
Full answer, verbatim: “The launch announcement says Free users get GPT-6 Luna; the Help Center says GPT-5.6 Luna. That is a genuine discrepancy in OpenAI’s documentation—not evidence, by itself, that the two names refer to the same model. What each source says: October 7 GPT-6 launch announcement | GPT-6 Luna | The rollout began for paid plans on October 7 and was to expand to Free and Go starting October 8. Paid Chat plans received GPT-6 Sol. October 7 ChatGPT release notes | GPT-6 Luna | These independently repeat: “Plus, Pro, Business, and Enterprise use GPT-6 Sol; Free and Go use GPT-6 Luna.” Help Center model-availability article | GPT-5.6 Luna | It says Free and Go’s default is GPT-5.6 Luna, that Instant uses it, and that their Think button also uses GPT-5.6 Luna—not Sol. One qualification about your October 22 cutoff: I verified the live Help Center article, which still contains that GPT-5.6 Luna wording, but I did not find a dated October 22 snapshot. I can therefore confirm the conflicting wording, but cannot independently certify the article’s exact historical state on October 22. Why do they differ? The supported explanation is that the launch materials and the operational Help Center article are not synchronized. The announcement describes the GPT-6 rollout to Free and Go, while the Help Center combines new GPT-6 guidance for eligible paid accounts with repeated GPT-5.6 Luna guidance for Free and Go. That makes documentation lag or an incomplete update a plausible explanation—but the sources do not explicitly establish the cause. It would go beyond the evidence to claim that OpenAI officially renamed GPT-5.6 Luna, rolled Free users back, or deliberately kept all Free users on the older model. A phased rollout could explain different account experiences, but does not by itself resolve the contradictory model names in the documents. Neither is this simply a Chat-versus-Work distinction: the launch announcement explicitly applies to Chat and says Work and Codex were unchanged. The careful way to report it is: “OpenAI announced GPT-6 Luna for Free and Go in Chat, but its Help Center continued to document GPT-5.6 Luna for those plans; the official sources conflict, and no verified explanation reconciles them.””
Score: correct — met the multi-hop-conflict requirement (gave BOTH, noted the conflict, qualified the 22 October cutoff).
Run A4 — Perplexity × Prompt 4 (finished 2:02 AM PDT)
Model shown: none. Inline citations: yes, 9 chips (25 sources button), to the Perplexity Help Center plan-comparison article and OpenAI Deep Research / Pro-tier help pages.
Full answer, verbatim: “The price winner is Perplexity Pro—but its adequacy for 30 Deep Research reports a month cannot be established from the current published limits. Nor do OpenAI’s current public documents give numeric Deep Research allowances for ChatGPT Plus or Pro $100. A definitive “cheapest adequate tier” therefore needs the actual account allowance, not just the subscription price. The maths: Treat the requirement as 30 completed Research-mode reports per 30-day period, with one research task producing one report and no retries. All prices below are USD, before tax. Required allowance = 1 × 30 = 30 reports. A plan qualifies only when its available Research allowance is at least 30; ordinary searches or chat messages do not count toward that requirement. ChatGPT Plus | $20/month | 20/30=$0.67 | No. OpenAI describes “expanded deep research,” but its current help page directs users to the in-product task counter rather than publishing a numeric plan quota. ChatGPT Pro $100 | $100/month | 100/30=$3.33 | No. The $100 tier exists, but the current documents reviewed do not specify its numeric Deep Research allowance. Perplexity Pro, monthly billing | $20/month | 20/30=$0.67 | No. Its official comparison lists Research queries as “Monthly limits (average use),” without a number. Perplexity Pro, annual billing | $200/year; 200/12=$16.67/month equivalent | 200/(12×30)=$0.56 | Same unquantified Research limit; annual billing changes the price, not the evidence of capacity. These per-report figures are conditional costs at the required output—not proof that each plan can deliver that output. What would make each plan the cheapest adequate choice? If Lᴾ ≥ 30, Perplexity Pro annually is cheapest: $200 for the assumed 360 reports per year, or about $0.56/report. If both Lᴾ ≥ 30 and L₊ ≥ 30, monthly Perplexity Pro and ChatGPT Plus tie at $20/month, or about $0.67/report. If both $20 options fall below 30 but L₁₀₀ ≥ 30, ChatGPT Pro $100 is the cheapest adequate option among these three: $3.33/report. If none has an allowance of at least 30, none qualifies as a standalone subscription for this workload. The $100 plan costs 100/20=5 times as much as either monthly $20 option. That price ratio does not establish a fivefold Research quota; OpenAI’s current documentation does not provide that numerical relationship. Why a confident quota-based answer would be misleading: OpenAI currently says that Deep Research usage varies by plan and that the account’s in-product counter shows remaining tasks. Fixed monthly allowances reset every 30 days from first use—not necessarily on the calendar-month boundary. Perplexity’s current official table distinguishes Pro Searches from Research queries and leaves the consumer Pro Research number unspecified. Its published 50 Research queries/month applies to Enterprise Pro, not consumer Pro; substituting that number would compare the wrong tier. For example, if an account actually shows a 25-report allowance, the calculation is: 30−25=5 reports short, 25/30=83.3% coverage. That plan would be inadequate regardless of its attractive nominal cost per included report. But 25 must be verified for the account—it should not be presented as a current official quota from the documents above. Purchase conclusion: Perplexity Pro at $200/year is the lowest-priced candidate, saving 240−200=$40 annually against either $20/month option. Buy it for this requirement only after confirming that its Research allowance covers at least 30 reports per 30 days. On the current public evidence, calling any of these three the definitively cheapest adequate tier would overstate what is known.”
Score: correct — maths checks out (20/30=$0.67, 100/30=$3.33, 200/(12×30)=$0.56, 240−200=$40), and the required caveat (adequacy unproven on published evidence) was flagged.
Run A5 — Perplexity × Prompt 5 (finished 2:06 AM PDT)
Model shown: none. Inline citations: yes, 5 chips (11 sources button), to OpenAI ChatGPT release notes and the About ChatGPT Pro tiers article.
Full answer, verbatim: “On 29 October 2026, OpenAI increased ChatGPT Pro’s Work and Codex allowances and cancelled a previously announced Pro 200 reduction. Plus usage was unchanged. This was not a blanket doubling of every ChatGPT limit. What changed by Pro tier: Pro 100 — $100/month | Doubled from its previous allowance. Pro 200 — $200/month, previously on the lower allowance | Doubled to match the existing higher Pro 200 allowance. Pro 200 — $200/month, already on the higher allowance | Kept that allowance; it did not double again. The reduction announced by email on 29 September was cancelled. Pro 500 — $500/month | Doubled from its previous allowance. The dated release notes establish these changes; OpenAI’s Pro-tier documentation confirms the prices and the distinction between the two previous Pro 200 allowances. Existing subscribers receive the updated allowance automatically, without waiting for renewal or cancelling and resubscribing. New subscriptions include the updated allowance. Pro 200 remains $200/month, and the allowance change does not give it Ultrafast access—that remains exclusive to Pro 500. Separate Chat message limits: OpenAI distinguishes Work and Codex usage from the weekly GPT-6 Pro message limit in Chat: Pro 100 and Pro 500 retain their existing weekly Chat limits. Pro 200’s lower weekly Chat limit rises to match the higher one; subscribers already on the higher limit keep it. Did Plus change? No—not in this update. OpenAI explicitly states, “Plus usage is unchanged.” It also cautions that doubling an allowance does not guarantee twice as many completed tasks: consumption depends on the model, task and settings.”
Score: correct — matched the expected 29 October facts.
Run B1 — ChatGPT default × Prompt 1 (finished ~2:08 AM PDT)
Model shown: none. Inline citations: yes, 3 linked sources in text. Citations in order: Perplexity Help Center plan comparison (Free: 3 Pro Searches/day; 1 Research query/month; updated September 2, 2026); Perplexity pricing page (described as suitable for limited daily usage, no exact number — note: the link used pointed to Perplexity’s ENTERPRISE pricing page, not the consumer Pro pricing page); The Tool Sweep’s Perplexity review as the third-party source of the 5/day claim.
Full answer, verbatim: “The correct figure is 3 Pro Searches per day on Perplexity Free, according to Perplexity’s own Help Center (updated September 2, 2026). The claim of 5 searches per day appears in some third-party reports, but it conflicts with Perplexity’s current official plan comparison. There is also an important distinction between the pricing page and the Help Center. What the sources say: Perplexity Help Center | 3/day. Perplexity pricing page | No exact number. Third-party report claiming 5/day | 5/day, conflicting information. Official Help Center: Perplexity’s plan comparison explicitly lists: Free: 3 Pro Searches/day; Free: 1 Research query/month; Basic searches: practically unlimited. Official pricing page: The Perplexity pricing page describes Free as suitable for limited daily usage. It does not specify 3 or 5 Pro Searches. Where does the 5-search figure come from? A third-party article, The Tool Sweep’s Perplexity review, reports five daily searches, attributing that number to another Help Center article. However, its claim conflicts with the current official plan comparison. Final verdict: 3 Pro Searches/day: Best-supported current figure. 5 Pro Searches/day: Conflicting, not confirmed by the current plan comparison. Does the official pricing page publish the number? No. Does Perplexity publish the number elsewhere officially? Yes, in its Help Center. For an accurate comparison of free AI tools, I’d use 3 daily Pro Searches and cite Perplexity’s Help Center rather than its pricing page.”
Score: correct.
Run B2 — ChatGPT default × Prompt 2 (finished ~2:10 AM PDT)
Model shown: none. Inline citations: yes, 2 (1 inline link + 1 citation-button expansion), to OpenAI pricing and Deep Research help pages.
Full answer, verbatim: “ChatGPT Go includes Deep Research because OpenAI is making advanced AI research more accessible at a lower price, rather than reserving the feature exclusively for Plus subscribers. However, there is an important distinction: having Deep Research access does not mean having Plus-level Deep Research usage. Go vs Plus: What’s the difference? According to OpenAI’s official pricing page: Deep Research | Included | Included; Usage allowance | Limited | Expanded; Advanced reasoning models | More limited | Expanded access; Projects and scheduled tasks | More limited | Included; Overall productivity features | Basic expanded access | More advanced. Why would OpenAI do this? There are three likely business reasons: 1. Competition: Other AI services offer research capabilities at relatively affordable prices. OpenAI benefits from making its research tools accessible to more people. 2. Upselling: Go introduces users to advanced features. When users need more research tasks, Plus becomes a natural upgrade. 3. Product differentiation: OpenAI can offer the same underlying feature across different subscriptions while differentiating plans through usage limits, models and additional tools. These are business interpretations, not officially confirmed explanations from OpenAI. Does Go have the same number of research tasks as Plus? No. OpenAI describes Go’s Deep Research access as limited and Plus’s as expanded. Interestingly, OpenAI does not publish a fixed numerical allowance for these plans on its pricing page. Its Help Center says usage varies by plan and that the in-product counter shows remaining tasks. My conclusion: Go gives you access to the Deep Research capability, while Plus sells you more capacity and a broader set of advanced features. So, Go is not equivalent to Plus, even when both offer Deep Research.”
Score: correct — the opening sentence explains the false premise as if true, BUT it is immediately corrected with citation (“does not mean having Plus-level Deep Research usage”; “Go is not equivalent to Plus”) and the speculative “why” section is explicitly labelled “business interpretations, not officially confirmed”. That hedge/cite exception is why it scores correct, not wrong-and-would-miss. Readers should still note the opening line alone, out of context, would mislead.
Run B3 — ChatGPT default × Prompt 3 (finished ~2:11 AM PDT)
Model shown: none. Inline citations: yes (2 inline links + 3 citation buttons), to the GPT-6 launch announcement, release notes, and Help Center model article.
Full answer, verbatim: “There is an important distinction here: the October 7 launch announcement and the October 22 Help Center documentation describe different things at different levels of specificity. The launch announcement explicitly identifies the model powering ChatGPT Free. The Help Center is a separate source describing model availability and plan restrictions, and its wording can change as OpenAI updates the documentation. 1. What the October 7 launch announcement says: Free and Go: GPT-6 Luna. Plus, Pro, Business and Enterprise: GPT-6 Sol. Free and Go rollout began October 8, following the paid rollout on October 7. This is explicit: Free users were included in the GPT-6 launch, but received Luna rather than Sol. 2. What the Help Center says: There is a significant documentation discrepancy. The Help Center’s currently retrieved article, GPT-6 and other models in ChatGPT, now says: Free and Go users receive GPT-5.6 Luna. They do not have access to GPT-5.6 Sol. Think also uses GPT-5.6 Luna. However, the official October 7 release notes independently repeat the original announcement: Free and Go use GPT-6 Luna. 3. Why do they differ? There are three possible explanations, but the evidence supports a documentation inconsistency more strongly than a confirmed model downgrade. Different documentation timelines: The Help Center article has been updated since October 22. The version currently visible says GPT-5.6 Luna, so it cannot automatically be treated as the October 22 wording. Different model generations: GPT-5.6 Luna and GPT-6 Luna are different model versions. The shared name Luna does not make them identical. Documentation lag or error: The October 7 announcement and October 7 release notes explicitly identify GPT-6 Luna for Free users. The current Help Center’s GPT-5.6 Luna wording conflicts with that announcement. 4. Final verdict: October 7 launch announcement | GPT-6 Luna; October 7 release notes | GPT-6 Luna; Currently retrieved Help Center | GPT-5.6 Luna. My conclusion: OpenAI’s launch announcement clearly promised GPT-6 Luna to Free users. The Help Center now says GPT-5.6 Luna, but I cannot establish from the available historical evidence that this was also what it said specifically on October 22. I would not interpret the discrepancy as proof that Free users were downgraded.”
Score: correct — gave both sides, noted the genuine discrepancy, qualified the 22 October cutoff.
Run B4 — ChatGPT default × Prompt 4 (finished ~2:11 AM PDT)
Model shown: none. Inline citations: yes (3 inline links + 4 citation buttons), to OpenAI pricing, Pro-tier documentation, Deep Research help, and the Perplexity Help Center comparison.
Full answer, verbatim: “Cost comparison: 30 Deep Research reports per month. Pricing and published limits checked: 31 October 2026 · USD · Individual subscriptions. The researcher needs: 1 completed Deep Research report every day. 30 reports per 30-day month. A plan that can reliably accommodate all 30, without running out halfway through the month. There is an important distinction between Deep Research reports and ordinary searches or Pro Searches. They are not interchangeable. Monthly cost comparison: ChatGPT Plus | $20 | Insufficient at 25 tasks; ChatGPT Pro 100 | $100 | Highest ChatGPT research allowance; Perplexity Pro | $20 | Research allowance not numerically guaranteed publicly. ChatGPT Plus: $20/month. OpenAI’s published Deep Research allowance was: 25 tasks per month for Plus. 250 tasks per month for Pro. These figures were explicitly published in OpenAI’s April 24, 2025 announcement. Using the 25-task allowance: 30−25=5 missing reports; 25/30×100=83.33%. So Plus covers only 83.33% of the researcher’s monthly requirement under that published allowance. Cost per available report: $20/25=$0.80. Verdict: Not adequate for 30 reports if the 25-task allowance applies. One qualification: OpenAI’s October 2026 Help Center now says limits vary by plan and directs users to their in-product usage counter. It does not confirm that the historical 25-task figure remains the current Plus allowance. ChatGPT Pro 100: $100/month. OpenAI currently offers a $100/month Pro tier. Its pricing page advertises maximum Deep Research, while the Pro documentation confirms that Pro 200 has more usage than Pro 100. The historical published Pro allowance was 250 tasks per month. If that allowance applies: 250−30=220 spare reports; 250/30=8.33× required capacity. Cost per required report: $100/30=$3.33. However, there is an important caveat: OpenAI’s current documentation does not publish a guaranteed numerical allowance specifically for the new Pro 100 tier. Therefore, 250 should not be treated as a verified October 2026 entitlement. Verdict: The strongest ChatGPT candidate, but confirm the actual account allowance before relying on it for a daily research workflow. Perplexity Pro: $20/month. Perplexity’s official pricing is: $20/month on monthly billing. $17/month equivalent when billed annually ($204 upfront). Its September 2, 2026 Help Center comparison distinguishes between: 1. Pro Searches. 2. Research Queries. For Perplexity Pro, Research Queries have monthly limits described as average use. The company does not publish a specific numerical Pro allowance in that comparison. This distinction matters: 3 Pro Searches per day does not mean 90 Deep Research reports per month. The free plan’s 3/day figure relates to Pro Searches, not Deep Research reports. For the paid Pro plan: $20/30=$0.67. But that is only the cost per required report if the plan actually permits 30 Research Queries. Verdict: Potentially the cheapest option, but 30 reports per month are not guaranteed by its published Pro limits. Final financial comparison: Monthly cost | $20 | $100 | $20; Published numerical research allowance | 25* | 250* | Not specified; Required | 30 | 30 | 30; Shortfall | 5* | 0* | Unknown; Cost per required report | — | $3.33 | $0.67**; Adequacy guaranteed by current published numerical limit? | No | No | No. *Historical OpenAI allowances, not verified current tier entitlements. **Conditional calculation assuming 30 reports are available. My recommendation: Best candidate for a guaranteed-capacity test: ChatGPT Pro 100. Why? Plus historically falls short by five reports. Perplexity Pro is cheaper, but its published research allowance does not establish 30 reports monthly. ChatGPT Pro offers the highest Deep Research allowance among the three options considered. However, I would not claim that Pro 100 is mathematically guaranteed to provide 30 reports based on the current public documentation. The honest conclusion: Cheapest advertised subscription: $20/month (Plus or Perplexity Pro). Cheapest subscription with a historically documented capacity above 30: ChatGPT Pro. Cheapest currently verified subscription that guarantees 30 reports: cannot be established from the published numerical limits. One practical next step: Check the Deep Research remaining-task counter on a Pro 100 account. That will establish whether the actual allowance is at least 30 in its current billing cycle.”
Score: correct — maths checks out, historical 25/250 figures explicitly footnoted as “not verified current tier entitlements”, and the required caveat (no tier’s 30-report adequacy can be established from current published numbers) was flagged. Its “best candidate” recommendation is conditional, not a definitive cheapest-adequate pick.
Run B5 — ChatGPT default × Prompt 5 (finished ~2:12 AM PDT)
Model shown: none. Inline citations: yes (3 official-source links + 5 citation buttons), to OpenAI ChatGPT release notes, About ChatGPT Pro tiers, and Managing usage with GPT-6 Astra in Work and Codex.
Full answer, verbatim: “ChatGPT Pro changes on 29 October 2026. Officially confirmed by OpenAI. On 29 October 2026, OpenAI increased Work and Codex usage allowances across its Pro plans and cancelled a previously announced reduction for Pro 200 subscribers. The changes did not include an increase to ChatGPT Plus allowances. 1. What changed for Pro? Pro 100 ($100/month) | Work and Codex allowance doubled. Pro 200 ($200/month) | Planned reduction cancelled; lower-allowance subscribers received double their previous allowance. Pro 500 ($500/month) | Work and Codex allowance doubled. Important distinction for Pro 200: Subscribers already receiving the higher allowance kept it. Subscribers on the lower allowance received twice as much, bringing them up to the same limit. They did not receive another doubling on top of the higher allowance. 2. Did ChatGPT Plus change? No announced Plus allowance increase on 29 October. Plus remained at its existing Work and Codex allowance. The October 29 announcement specifically names Pro 100, Pro 200, Pro 500 and Business Premium, not Plus. Plus still has limited GPT-6 Astra usage in Work and Codex, whereas Pro subscribers have substantially higher allowances. 3. What exactly was doubled? This is particularly important: ChatGPT Work usage. Codex usage. These two features share an allowance. It does not mean every ChatGPT feature received twice as much usage. For example, GPT-6 Pro messages in regular Chat have separate limits. Deep Research and other feature allowances should not be assumed to have doubled. 4. Official sources: 1. OpenAI – ChatGPT Release Notes — October 29 announcement. 2. OpenAI – About ChatGPT Pro Tiers — Pro prices and allowances. 3. OpenAI – Managing usage with GPT-6 Astra in Work and Codex — how shared usage works. Bottom line: OpenAI improved Pro without increasing subscription prices. Pro 100 and Pro 500 received doubled Work/Codex allowances, Pro 200’s planned reduction was cancelled and lower-allowance users were brought up to the higher level. Plus received no corresponding increase. One important qualification: OpenAI does not express these doubled allowances as a guaranteed number of research reports or coding tasks. Actual consumption depends on the model, task complexity and settings.”
Score: correct — matched the expected 29 October facts.
Run C1 — ChatGPT search-on × Prompt 1 (finished ~2:13 AM PDT)
Model shown: none. Inline citations: yes (1 citation button + 1 inline link + 1 citation button), to the Perplexity Help Center plan comparison and the pricing page (again, the enterprise pricing URL was used as the “pricing page”).
Full answer, verbatim: “Confirmed: 3 per day. As of 31 October 2026, Perplexity Free provides 3 Pro Searches per day, not 5. The confusion comes from older Perplexity documentation and guides that still mention five searches. What Perplexity officially says: Perplexity’s Help Center article, Which Perplexity Subscription Plan is right for you?, was updated on 2 September 2026. Its plan comparison explicitly lists: Free: 3 Pro Searches per day. Pro: Weekly limits, based on average use. Free Research: 1 Research query per month. This is the strongest source for the current allowance. Where does the 5-search figure come from? Perplexity’s own older Getting Started with Perplexity guide says: Free users can use Pro Search five times every four hours. That is likely the source of the conflicting information. However, it conflicts with the newer Help Center plan comparison. Does Perplexity publish the number on its pricing page? No. Perplexity’s official Pricing page describes Free as suitable for limited daily usage but does not specify the exact number of Pro Searches. The exact figure is published in its Help Center instead. Final comparison: 3 Pro Searches/day | Current official limit; 5 Pro Searches every 4 hours | Older documentation; Exact number on pricing page | Not published; Research queries on Free | 1 per month. My recommendation: If you’re comparing Perplexity Free against ChatGPT Free, use 3 Pro Searches daily in your calculations, not 5.”
Score: correct. Search-on vs default comparison (same prompt): explicit web search did NOT change the answer’s verdict. Both runs concluded 3 Pro Searches/day, cited the same Help Center article (updated 2 Sept 2026), and agreed the pricing page publishes no number. Differences were cosmetic: the search-on run framed the finding “As of 31 October 2026” and traced the “5” figure to Perplexity’s own older “Getting Started with Perplexity” guide (“five times every four hours”), while the default-mode run traced it to a third-party review (The Tool Sweep) and added a “Final verdict” bullet list.
Test anomalies — reported, not hidden
1. ChatGPT’s runs twice linked Perplexity’s “pricing page” to the ENTERPRISE pricing page (https://www.perplexity.ai/enterprise/pricing) rather than the consumer Pro pricing page — the wrong URL for the claim being made, even though the “publishes no number” conclusion still held. 2. Perplexity’s Prompt 1 answer said its Help Center article was “last modified October 29, 2026” while both ChatGPT runs said “updated September 2, 2026” — a minor internal inconsistency between the tools’ citations. 3. Neither tool displayed a model name in these free-tier default UIs; this was recorded as “not shown” rather than guessed. 4. No sign-in, payment, or new-account event occurred in any leg; both already-signed-in free accounts worked throughout. Nothing could not be run.
What the test actually proves — and what it does not
It proves both free tools, on these 5 hard prompts on 31 October 2026, handled conflict traps, a false premise, a multi-hop conflict, cost maths and a recent allowance change without producing a “wrong and you would have missed it” claim. It does NOT prove either tool is universally accurate, and it does NOT prove ChatGPT default is as good as search-on in general — on Prompt 1, search-on made no difference to the verdict, but that is one prompt, not a general finding. The article’s verdicts above rest on the pricing, model and accuracy research, not on this 11-run test alone.
Sources and method
- Perplexity pricing: https://www.perplexity.ai/hub/pricing — read 31 October 2026.
- ChatGPT pricing: https://chatgpt.com/pricing/ — read 31 October 2026.
- Perplexity free limits, conflicting: Fello AI (15 October 2026, 3 Pro/day, 1 Research/month) vs AI Price Compare (28 October 2026, 5 Pro/day) vs ToolColumn (Free 3/day, 1/month; Pro/Max numbers unpublished).
- ChatGPT Pro allowance update, 29 October 2026: secondary summary of an OpenAI help update (Pro $100/$500 Work/Codex doubled, Pro $200 restored, Plus unchanged).
- GPT-6 rollout and Free/Go conflict: OpenAI launch announcement (7–8 October 2026) vs Help Center as of 22 October 2026.
- Tow Center / Columbia Journalism Review, “AI Search Has a Citation Problem”, 6 March 2025, 1,600 queries.
- Deep Research URL-validity preprint: arXiv:2604.03173 (2026, not peer-reviewed; URL validity only, no Perplexity Deep Research figure).
- Perplexity model picker snapshot: Help Center via secondary sources, September 2026 (rotates — dated).
- Prices checked 31 October 2026. Free and paid limits that neither vendor publishes as numbers are labelled as secondary/conflicting in the text above — re-check them in-account at publish time.

1 comments