Perplexity Deep Research vs ChatGPT Deep Research: AI Research Agent Head-to-Head
Two agentic research modes at the same $20 Pro price. We compared them on benchmark accuracy, citation reliability, runtime, quotas, and free-tier access to decide which one belongs in a research workflow.
ChatGPT Deep Research takes the overall by two points, and one number carries most of the weight: 26.6% on Humanity's Last Exam versus Perplexity's 21.1%. It also produces the longer, more structured briefs and now supports MCP connectors and source scoping. Perplexity Deep Research wins on speed (2 to 4 minute reports versus 5 to 30 minutes), citation reliability, and free-tier access. It's the only tool of the two that ships Deep Research on a $0 plan. If you're writing long, polished analyst-style reports, ChatGPT is the higher-scoring pick. If you're doing fact-dense, citation-verifiable research on live topics, or you can't or won't pay $20/month, Perplexity is the better tool.
Perplexity and OpenAI shipped their deep-research modes within days of each other in February 2025 and have iterated aggressively since. Both are agentic: you submit one question, the tool writes its own research plan, browses dozens to hundreds of pages, and returns a cited report. Both cost the same $20/month on the entry paid tier. The buying decision isn't about price anymore. It's about which agent produces better measured results on the work a researcher actually does.
Every round below names the concrete procedure behind it. Accuracy rounds are scored on published, third-party benchmarks (Humanity's Last Exam, the Columbia Journalism Review citation audit, LMSYS factual-accuracy evaluation). Runtime, quota, and pricing rounds are pure measurement against each vendor's documentation as of the test date.
| Test category | Winner | Result & method |
|---|---|---|
| Reasoning benchmark accuracy | ChatGPT Deep Research | ChatGPT Deep Research posts the higher score on the most-cited third-party evaluation. On Humanity's Last Exam, Perplexity Deep Research scored 21.1% and ChatGPT's Deep Research scored 26.6%, both comfortably ahead of standalone models like GPT-4o (3.3%), Claude 3.5 Sonnet (4.3%), and Gemini Thinking (6.2%). The 5.5-point margin is small in absolute terms but consistent across independent reporting. How we measured it: Compared each tool's published score on Humanity's Last Exam, a 3,000+ question benchmark spanning 100+ subjects designed to stress-test frontier research systems, using each vendor's reported result. |
| Citation reliability | Perplexity Deep Research | Perplexity retrieves before it writes. ChatGPT synthesizes and then attaches citations. That architectural difference shows up in audits. CJR measured a 37% citation hallucination rate for Perplexity against 67% for ChatGPT Search, and an April 2026 LMSYS evaluation put Perplexity Pro at 92% factual accuracy on real-time queries versus ChatGPT at 87% with browsing enabled. The gap widened on fast-moving data (94% vs 81% on financial queries). How we measured it: Compared published third-party audits of citation behavior, including the Columbia Journalism Review citation audit and an April 2026 LMSYS factual-accuracy evaluation on real-time queries. |
| Runtime and throughput | Perplexity Deep Research | A single Perplexity Deep Research query runs dozens of searches, reads hundreds of sources, and returns a cited report in 2 to 4 minutes. ChatGPT Deep Research is documented at 5 to 30 minutes per report, browsing for up to 30 minutes on the heavier end. For a briefing before a meeting or an intraday research pass, the runtime gap is decisive. For an overnight literature review, it's not. How we measured it: Compared each vendor's documented and reported end-to-end runtime for a full Deep Research query on the current default model. |
| Report depth and structure | ChatGPT Deep Research | ChatGPT Deep Research produces the longer, more analytical brief, closer to what a human research analyst would write. Independent testing consistently describes ChatGPT's output as more structured and executive-ready, and characterizes Perplexity as a fast research briefing rather than a long-form report. The tradeoff is direct: ChatGPT's 5 to 30 minute browsing window buys more source layers before writing. How we measured it: Compared published reports from independent evaluators on structure, length, and analytical layering, and cross-referenced with each vendor's documented output length and browsing depth. |
| Free-tier access | Perplexity Deep Research | Perplexity's Free tier includes Deep Research, roughly 3 to 5 queries per day depending on the account. ChatGPT's Free and Go plans don't include Deep Research at all; the feature requires Plus ($20/month), Pro ($100 or $200), Business, or Enterprise. For students, journalists, and anyone evaluating the tools before buying, this round isn't close. How we measured it: Compared what each vendor makes available on its published $0 plan, per each vendor's pricing page as of the test date. |
| Paid quota | Perplexity Deep Research | At the same $20 price, Perplexity Pro currently documents around 20 Deep Research queries per day, and ChatGPT Plus documents 10 to 25 per month, with overflow dropping to a lightweight variant. Perplexity's cap has tightened materially in 2026, since earlier Pro accounts saw much higher daily allowances, but even at 20/day it substantially outstrips a monthly cap in the low double digits. How we measured it: Compared documented Deep Research allowances on each vendor's $20 Pro plan as of the test date. |
| Sourcing and workflow controls | ChatGPT Deep Research | In February 2026, OpenAI shipped a GPT-5.2-based Deep Research with source scoping (restricting to selected sites), MCP-server data connections, better steering, and real-time interruption. Perplexity offers Focus modes (Academic, Reddit, YouTube, Wolfram Alpha) and Spaces for private files, and Max users get Model Council orchestration across Claude Opus 4.6, GPT-5.2, and Gemini 3 Pro. Cross-source scoping and MCP tip the workflow-controls round to ChatGPT on the $20 tier. Model Council is a Max-tier ($200) advantage, not a Pro-tier one. How we measured it: Audited each vendor's shipped controls for restricting scope, connecting private data, and integrating with downstream workflows, per each vendor's official documentation as of the test date. |
Perplexity Deep Research and ChatGPT Deep Research are sold for the same job: turn one question into a cited, multi-source report without a human doing the browsing. They now charge the same $20 on the entry paid tier, so the comparison reduces to which agent produces better measured results.
Reading the result
The overall margin is two points. ChatGPT took three of seven rounds (reasoning benchmark accuracy, report depth, and workflow controls) and Perplexity took four (citation reliability, runtime, free-tier access, and paid quota). That split is the story. On the benchmark that most researchers cite, ChatGPT is measurably stronger. On the operational metrics that decide whether a tool gets used every day, Perplexity is stronger.
How to map the rounds to a buying decision
If the deliverable is a long, structured report you plan to publish or present, ChatGPT Deep Research is aimed at long, polished research briefs where the AI sounds like a human analyst: industry deep-dives, literature-style writeups, and reports you plan to export or share . The 5 to 30 minute browsing window and the higher HLE score both point the same way.
If the deliverable is a fast, cited briefing you need to verify (market research, competitive analysis, or anything with current prices, policies, or regulations), Perplexity is best for fact-dense topics where you need traceable citations in a hurry: market research, competitive analysis, news-adjacent questions, and anything involving current prices, policies, or regulations . The 2 to 4 minute runtime and the higher citation-accuracy audit results both point the same way.
On the benchmark gap
The 5.5-point Humanity’s Last Exam gap is the single most-cited data point in this comparison, and it deserves context. On Humanity’s Last Exam, a test of 3,000+ questions across 100+ subjects, Perplexity Deep Research scores 21.1% accuracy, trailing ChatGPT’s Deep Research (26.6%) but comfortably beating standalone models like GPT-4o (3.3%), Claude 3.5 Sonnet (4.3%), and Gemini Thinking (6.2%). Perplexity achieves this while completing most tasks in under 3 minutes, faster than ChatGPT’s 5-30 minute range.
Both scores dwarf standalone frontier models, and that’s the more important read: the agentic loop matters more than the underlying model. The gap between the two deep-research systems is smaller than the gap between either of them and a bare-metal LLM.
On citation reliability
The two systems handle citations very differently. Perplexity’s citations are persistent and numbered, with every claim in the response linking to a specific source you can click through and read yourself. This is genuinely useful for academic or professional research where you need to verify everything. ChatGPT in standard chat doesn’t cite; only Deep Research and browsing surface links, and independent audits find them less consistently integrated.
Neither tool is a substitute for verification. Citations don’t equal accuracy: an independent analysis by the Tow Center found Perplexity still answered incorrectly about 37% of the time despite citing sources, and Stanford researchers found it fabricated references roughly 26% of the time (versus 40% for ChatGPT). Both tools require the researcher to click through and confirm.
On the model bets
The two products have made different bets on the model underneath. Since February 2026, ChatGPT Deep Research uses a model based on OpenAI’s GPT-5.2, having originally been released with a specialized version of OpenAI’s o3. Perplexity has taken the model-agnostic route: Claude Opus 4.6 powers Deep Research on Perplexity, upgraded in February 2026 for Max users, with Pro rollout following.
That routing choice is why Perplexity’s benchmark score is close to ChatGPT’s despite Perplexity not owning a frontier model. It also explains the reliability profile: Perplexity’s model-agnostic approach means users are never locked into one model family, and when Claude Opus 4.6 gets upgraded, Deep Research workflows upgrade automatically without changing a setting.
On price parity, and where it breaks
At $20/month the two products list identically. Where the price picture diverges is the free tier and the top tier.
On the free tier: Perplexity’s free tier includes real-time web search and up to three Deep Research queries per day; ChatGPT’s free tier has no web access. For students and budget-constrained researchers, that makes Perplexity the practical default starting point.
On the top tier, both vendors list a $200/month plan, but they buy different things. Perplexity Max at $200/month matches ChatGPT Pro in price but takes a different approach: 19-model orchestration including GPT-5.2 and Claude Opus 4.6, versus ChatGPT Pro’s single flagship model depth.
ChatGPT Pro subscribers ($200/month) receive 250 Deep Research queries per month (half of which are “lightweight”), Plus, Team and Enterprise users receive 25 queries per month, and free users receive 5 “lightweight” queries per month. A team writing dozens of long reports a month will find the $200 ChatGPT tier easier to justify than the $200 Perplexity tier. A team running many parallel comparative queries will find Perplexity Max’s Model Council architecture more useful.
On the recent product trajectory
Both products shipped material updates in early 2026 that any 12-month buying decision should price in. In February 2026, OpenAI announced updates to Deep Research with a new GPT-5.2-based model, better steering, limiting scope to select sites, connecting additional data using MCP servers, and better UI for the final reports. Perplexity has shipped in the same window: In March 2026, multimodal Deep Research shipped, adding the ability to analyze images and visual data within research workflows.
The direction of travel is the same on both sides: more sources, more scoping, more downstream integration. What isn’t converging is the operational profile. ChatGPT is still the slower, deeper, more polished agent. Perplexity is still the faster, more cite-verifiable one. A serious researcher will end up owning both.
- https://www.perplexity.ai/enterprise/pricing
- https://chatgpt.com/pricing/
- https://en.wikipedia.org/wiki/ChatGPT_Deep_Research
Priya Raman runs the Top AI Tracker test bench. She designs the scoring rubrics, sets the weightings for each category, and signs off on every published score. Her background is in systems evaluation and reproducible measurement.