Research Engine

Vital signs

Why it's on the table

On the table, Research Engine (Re) is seat 18 of 58, in the Knowledge & Memory family. It is a stable element — the category has settled, choosing is cheap, and switching is rare. Pick a holder and move on; this is not where your decision budget should go. It is marked essential: a company that leaves this seat empty is running with a gap that compounds. It sits in the lowest paid band — lunch money against the hours it returns.

Research Engine: the top 5 — v2026.Q3

Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.

1. 1PerplexityPerplexity AI

   Free · Pro $20/mo · Max $200/mo · Enterprise Pro $40/seat · usage-based agent credits on Max

   Best for The default daily research engine — fast, cited answers dozens of times a day, with Research mode and the Comet browser folding search into everything you do.

   The answer-engine category leader by every measure: $500M ARR (Apr 2026, up 335% YoY), $21.2B Series E-6 valuation (early 2026), 780M queries/month back in May 2025 and 100M+ MAU since. Comet — free since Oct 2025, iOS since Mar 2026 — turns the browser itself into the research surface, and the Feb 2026 subscription-first shift bet the company on answer quality over ads.

   Watch Valuation runs ~40x ARR against Google AI Mode giving cited answers away for free inside Search. Publisher scraping-and-attribution fights (News Corp et al.) are unresolved, and answer-level citation accuracy still demands spot-checking. Max/Enterprise pricing was 404 on the vendor page to our crawler — verified via Sacra only.

   $500M ARR (Apr 2026), +335% YoY from $232M 2025 revenue [src] · $21.21B valuation after Series E-6 (early 2026); subscription-first shift Feb 2026 [src] · Comet browser free for everyone Oct 2025; Android Nov 20, 2025; iOS Mar 18, 2026 [src]
   Site →Docs →

2. 2ChatGPT deep researchOpenAI

   In ChatGPT plans: Free (limited/lightweight) · Plus $20/mo · Pro $200/mo (maximum quota) · Business per-seat

   Best for The single exhaustive report — 5–30 minutes of autonomous browsing that replaces a day of reading, footnoted end to end.

   The mode that defined the deep-research category (Feb 2, 2025, 26.6% on Humanity's Last Exam vs o1's 9.1%) and the most heavily iterated since: rebuilt on a GPT-5.2-based model in Feb 2026 with steering, source limits, and MCP-server data connections. Distribution through ChatGPT's enormous user base makes it the deep-research mode most founders already have.

   Watch OpenAI itself warns it "occasionally makes factual hallucinations" and states rumors without conveying uncertainty. Quotas went opaque in 2026 — the pricing page now says only "limited/expanded/maximum"; the last published numbers (Jun 2025) were Pro 250/mo, Plus 25, Free 5 lightweight. Max quota costs $200/mo.

   Launched Feb 2, 2025; 26.6% Humanity's Last Exam vs o1 9.1% [src] · Feb 2026: migrated to GPT-5.2-based model; added steering, source limits, MCP integration [src] · Last published quotas (Jun 2025): Pro 250/mo, Plus/Team 25, Free 5 lightweight [src]
   Site →Docs →

3. 3Gemini Deep ResearchGoogle

   Free tier · AI Pro ~$20/mo (expanded access) · AI Ultra ~$250/mo (highest limits + report visuals)

   Best for Free-tier depth and research grounded in your own Gmail, Drive, and Chat alongside the open web.

   The distribution and grounding play: the original deep-research mode (Dec 2024), now Gemini 3-powered, live in 150+ countries and 45+ languages, and the only top pick that browses your Workspace data and hundreds of websites in one run, with audio overviews of the finished report. The free tier makes it the zero-budget default.

   Watch Numeric quotas are unpublished — 'higher limits' is all Google commits to. Report visuals (charts, simulators) are gated to the ~$250/mo Ultra tier and vanish entirely if you include Workspace sources. The 18+ requirement and Keep-Activity dependency for report history are easy to trip over.

   Available in 150+ countries, 45+ languages; Gemini 3-powered reports (2026) [src] · Browses hundreds of websites plus Gmail, Drive, and Chat with permission; 5–10 min per report [src] · Visuals exclusive to AI Ultra; unavailable when Workspace sources are included [src]
   Site →Docs →

4. 4Claude ResearchAnthropic

   Included in all paid Claude plans: Pro $17–20/mo · Max from $100/mo · Team $20/seat · Enterprise

   Best for Sourced investigation woven into the assistant where your work already lives — chained agentic searches over web plus MCP connectors, Workspace, and enterprise document catalogs.

   Launched Apr 15, 2025 as a Max/Team/Enterprise beta, now on every paid plan down to Pro. Its distinct approach — multiple searches that build on each other with easy-to-check citations, plus Gmail/Calendar/Docs integration and admin-enabled RAG cataloging on Enterprise — makes it the strongest pick for research entangled with your own context rather than the open web alone.

   Watch No dedicated quota: Research shares your plan's normal usage limits and burns them noticeably faster. Output favors verifiable synthesis over the hour-long, hundred-source reports ChatGPT and Gemini produce — a feature for checking, a limit for exhaustiveness.

   Launched Apr 15, 2025 (Max/Team/Enterprise beta, US/Japan/Brazil) with Google Workspace integration [src] · Now included on all paid plans (Pro and up), web/desktop/mobile (2026) [src] · Uses shared plan limits — "Research sessions can use up your limits faster" [src]
   Site →Docs →

5. 5ElicitElicit (spun out of Ought)

   Free (limited agent/reports) · Pro $49/mo · Scale $169/mo · Enterprise custom

   Best for When the question is scientific — literature search, screening, and data extraction over 138M papers with systematic-review rigor instead of open-web vibes.

   The scientific-literature research engine with real workflow depth: Research Agent, Reports, and automated Systematic Literature Reviews across 138M papers and 545k clinical trials, trusted by 5M+ researchers. Enterprise tier claims PRISMA-grade screening/extraction accuracy — the standard actual evidence teams are held to — with vendor-reported 80% time savings on reviews.

   Watch Vendor-claimed metrics (5M researchers, 80% savings) lack third-party audit. Literature-only — it will not research the open web, markets, or competitors. Serious use lands on Pro fast, and figure extraction is gated to the $169/mo Scale tier.

   138M papers + 545k clinical trials; 5M+ researchers (Aug 2026) [src] · Pro $49/mo, Scale $169/mo; Enterprise claims PRISMA-grade accuracy, 40k-paper screening [src] · Vendor-reported up to 80% time savings on systematic reviews [src]
   Site →Docs →

Research Engine: the top 8 compared

Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.

How to choose your research engine

If you ask dozens of sourced questions a day and want the fastest answer-with-citations loop

Perplexity Pro at $20 — and let Comet make the browser the research surface. At $0, Gemini's free tier is the fallback.

If the deliverable is one deep, footnoted report that replaces a day of reading

ChatGPT deep research — Plus covers most founders; only report-a-day power users need Pro's maximum quota at $200.

If half the answer lives in your own inbox, docs, and meetings

Gemini Deep Research with Workspace sources — or M365 Copilot Researcher in Microsoft shops, or Claude Research if your context is already wired up via MCP connectors.

If the question is scientific, clinical, or will be peer-reviewed

Elicit for the workflow (screening, extraction, PRISMA-grade reviews); Consensus for fast evidence checks; Undermind when discovery itself is the hard part.

If research is a feature of your product or agent, not a tab you open

Buy the API layer, not a subscription: Exa for index quality, Parallel for price-performance ($1/1k entry), Tavily for agent-framework ubiquity.

Research Engine: the whole field

19 more tools tracked in this category, including 4 dead, renamed, or sunsetting — a reference that hides the graveyard isn't one. Verified 2026-08-06.

Research Engine: the category in numbers

Edition v2026.Q3 · ranking, pricing and status verified 2026-08-06.

• Perplexity: $232M revenue 2025 → $500M ARR by Apr 2026 (+335% YoY), $21.2B Series E-6 valuation (early 2026), subscription-first pivot Feb 2026 [src]
• Deep research went from novelty (OpenAI, Feb 2025) to table stakes in ~12 months — by Feb 2026 every major assistant ships a mode, and OpenAI's was rebuilt on a GPT-5.2-base with MCP data connections [src]
• The agent-search API layer consolidated in Sept 2025–2026: Exa $85M at $700M (Benchmark), You.com $100M at $1.5B (Cox), Tavily $25M then absorbed as a Nebius subsidiary; Parallel opened a price war at $1/1k requests [src]
• Google AI Mode (Mar 2025 launch, global rollout through 2025) gives away cited AI answers inside Search — the structural squeeze on every standalone answer engine's free tier [src]
• Paid human search stays viable but niche: Kagi at 73,842 paying members and ~1.06M queries/day (Aug 6, 2026) [src]
• The scientific-research vertical is its own market now: Elicit claims 5M+ researchers, SciSpace 9.6M+, and 1,000+ GSK scientists run Undermind (2026) [src]

Research Engine: method & sources

Ranking criteria: speed-to-trustworthy-cited-answer, depth of autonomous research, grounding in the user's own data, and evidence discipline of the output — editorial picks, no affiliate consideration. Conflicts resolved: Perplexity's valuation — we cite Wikipedia's $21.21B Series E-6 (early 2026) over Sacra's $20B (Sept 2025) because it is the later event; both are secondary sources, as Perplexity publishes neither. Perplexity's Max ($200/mo) and Enterprise Pro ($40/seat) pricing could not be confirmed on perplexity.ai (pricing page 404 to our crawler) and rests on Sacra — re-check before quoting. Consensus's exact prices are omitted for the same reason (JS-rendered pricing page). ChatGPT deep-research quotas went deliberately vague in 2026 ('limited/expanded/maximum'); the numeric quotas cited are the last published (Jun 2025). Elicit states 5M+ researchers on its homepage but 2M+ on its pricing page — we used the homepage figure. Parallel's benchmark numbers are vendor-published and unaudited. Grok's DeepSearch branding no longer appears on x.ai; treated as folded into SuperGrok's agentic reasoning rather than dead. Adjacent elements: NotebookLM and knowledge bases → Kw · Knowledge Wiki; vector search infra → Vd; meeting transcripts → Mt; agent memory → Mm; Comet-style browser agents overlap the agent elements but are covered here only as Perplexity distribution. Search-for-agents APIs (Exa, Tavily, Parallel, Brave, You.com) sit on the Re/agent-infra boundary; kept here because the element scope names Exa explicitly. Ranking criteria: verified commercial traction, independent satisfaction surveys, agent benchmarks, and founder-fit (price floor, lock-in, surfaces). Editorial, never paid — the charter. Machine-readable twin: re.json.

Our take

Deep-research modes changed the job: you now delegate the reading, not just the searching. Keep the judgment.

Combines with

Appears in compounds

The Solo-Founder Compound

Guides

Best deep research tool for founders (2026)