{ "sym": "Re", "updated": "2026-08-06", "verdict": "Perplexity, for the daily habit \u2014 the category it invented is now a $500M-ARR, $21B business, and no rival matches its speed-to-cited-answer across free, Pro, and the Comet browser. But deep-research modes changed the job in 2025\u201326: for the one exhaustive report that replaces a day of reading, ChatGPT deep research (rebuilt on a GPT-5.2-base in Feb 2026) is the strongest single tool. Gemini Deep Research wins when the budget is $0 or the answers live in your Gmail/Drive; Claude Research wins when your work context already lives in Claude and MCP connectors; Elicit wins the moment the question is scientific literature rather than the open web. Keep the judgment \u2014 every one of these still hallucinates at the margins.", "top5": [ { "rank": 1, "name": "Perplexity", "maker": "Perplexity AI", "url": "https://www.perplexity.ai", "docs": "https://docs.perplexity.ai", "pricing": "Free \u00b7 Pro $20/mo \u00b7 Max $200/mo \u00b7 Enterprise Pro $40/seat \u00b7 usage-based agent credits on Max", "best_for": "The default daily research engine \u2014 fast, cited answers dozens of times a day, with Research mode and the Comet browser folding search into everything you do.", "why": "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 \u2014 free since Oct 2025, iOS since Mar 2026 \u2014 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 \u2014 verified via Sacra only.", "evidence": [ { "stat": "$500M ARR (Apr 2026), +335% YoY from $232M 2025 revenue", "src": "https://sacra.com/c/perplexity/" }, { "stat": "$21.21B valuation after Series E-6 (early 2026); subscription-first shift Feb 2026", "src": "https://en.wikipedia.org/wiki/Perplexity_AI" }, { "stat": "Comet browser free for everyone Oct 2025; Android Nov 20, 2025; iOS Mar 18, 2026", "src": "https://en.wikipedia.org/wiki/Comet_(web_browser)" } ], "tile_note": "cited answers, daily default" }, { "rank": 2, "name": "ChatGPT deep research", "maker": "OpenAI", "url": "https://openai.com/index/introducing-deep-research/", "docs": "https://help.openai.com/en/articles/10500283-deep-research-faq", "pricing": "In ChatGPT plans: Free (limited/lightweight) \u00b7 Plus $20/mo \u00b7 Pro $200/mo (maximum quota) \u00b7 Business per-seat", "best_for": "The single exhaustive report \u2014 5\u201330 minutes of autonomous browsing that replaces a day of reading, footnoted end to end.", "why": "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 \u2014 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.", "evidence": [ { "stat": "Launched Feb 2, 2025; 26.6% Humanity's Last Exam vs o1 9.1%", "src": "https://openai.com/index/introducing-deep-research/" }, { "stat": "Feb 2026: migrated to GPT-5.2-based model; added steering, source limits, MCP integration", "src": "https://en.wikipedia.org/wiki/Deep_research" }, { "stat": "Last published quotas (Jun 2025): Pro 250/mo, Plus/Team 25, Free 5 lightweight", "src": "https://en.wikipedia.org/wiki/Deep_research" } ], "tile_note": "deepest reports, biggest reach" }, { "rank": 3, "name": "Gemini Deep Research", "maker": "Google", "url": "https://gemini.google/overview/deep-research/", "docs": "https://support.google.com/gemini/answer/15719111", "pricing": "Free tier \u00b7 AI Pro ~$20/mo (expanded access) \u00b7 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.", "why": "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 \u2014 '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.", "evidence": [ { "stat": "Available in 150+ countries, 45+ languages; Gemini 3-powered reports (2026)", "src": "https://gemini.google/overview/deep-research/" }, { "stat": "Browses hundreds of websites plus Gmail, Drive, and Chat with permission; 5\u201310 min per report", "src": "https://support.google.com/gemini/answer/15719111" }, { "stat": "Visuals exclusive to AI Ultra; unavailable when Workspace sources are included", "src": "https://support.google.com/gemini/answer/15719111" } ], "tile_note": "free-tier depth, workspace-grounded" }, { "rank": 4, "name": "Claude Research", "maker": "Anthropic", "url": "https://claude.com/blog/research", "docs": "https://support.claude.com/en/articles/11088861-using-research-on-claude", "pricing": "Included in all paid Claude plans: Pro $17\u201320/mo \u00b7 Max from $100/mo \u00b7 Team $20/seat \u00b7 Enterprise", "best_for": "Sourced investigation woven into the assistant where your work already lives \u2014 chained agentic searches over web plus MCP connectors, Workspace, and enterprise document catalogs.", "why": "Launched Apr 15, 2025 as a Max/Team/Enterprise beta, now on every paid plan down to Pro. Its distinct approach \u2014 multiple searches that build on each other with easy-to-check citations, plus Gmail/Calendar/Docs integration and admin-enabled RAG cataloging on Enterprise \u2014 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 \u2014 a feature for checking, a limit for exhaustiveness.", "evidence": [ { "stat": "Launched Apr 15, 2025 (Max/Team/Enterprise beta, US/Japan/Brazil) with Google Workspace integration", "src": "https://claude.com/blog/research" }, { "stat": "Now included on all paid plans (Pro and up), web/desktop/mobile (2026)", "src": "https://support.claude.com/en/articles/11088861-using-research-on-claude" }, { "stat": "Uses shared plan limits \u2014 \"Research sessions can use up your limits faster\"", "src": "https://support.claude.com/en/articles/11088861-using-research-on-claude" } ], "tile_note": "agentic search, your context" }, { "rank": 5, "name": "Elicit", "maker": "Elicit (spun out of Ought)", "url": "https://elicit.com", "docs": "https://support.elicit.com", "pricing": "Free (limited agent/reports) \u00b7 Pro $49/mo \u00b7 Scale $169/mo \u00b7 Enterprise custom", "best_for": "When the question is scientific \u2014 literature search, screening, and data extraction over 138M papers with systematic-review rigor instead of open-web vibes.", "why": "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 \u2014 the standard actual evidence teams are held to \u2014 with vendor-reported 80% time savings on reviews.", "watch": "Vendor-claimed metrics (5M researchers, 80% savings) lack third-party audit. Literature-only \u2014 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.", "evidence": [ { "stat": "138M papers + 545k clinical trials; 5M+ researchers (Aug 2026)", "src": "https://elicit.com" }, { "stat": "Pro $49/mo, Scale $169/mo; Enterprise claims PRISMA-grade accuracy, 40k-paper screening", "src": "https://elicit.com/pricing" }, { "stat": "Vendor-reported up to 80% time savings on systematic reviews", "src": "https://elicit.com" } ], "tile_note": "science literature, systematic rigor" } ], "matrix": { "cols": [ "Pricing", "Deep-research quota", "Corpus", "Own-data grounding", "API", "Citations", "Best at" ], "rows": [ [ "Perplexity", "$0\u2013200/mo", "Research mode on Pro; agent credits on Max", "Live web + own index", "Enterprise file indexing (500 files)", "Search / Agent / Embeddings APIs", "Inline; 10x citations on Pro", "Daily cited answers" ], [ "ChatGPT deep research", "$20\u2013200/mo", "Pro max \u00b7 Plus expanded \u00b7 Free limited (opaque since 2026)", "Live web", "Connectors + MCP servers (Feb 2026)", "Deep-research models via API", "Footnoted long report", "One exhaustive report" ], [ "Gemini Deep Research", "$0 \u00b7 ~$20 \u00b7 ~$250", "Daily limits, numbers unpublished", "Live web", "Gmail \u00b7 Drive \u00b7 Chat", "No (app only)", "Linked sources + audio overview", "Free-tier breadth" ], [ "Claude Research", "$17\u2013100+/mo", "Shares plan limits, no dedicated quota", "Live web", "MCP connectors \u00b7 Workspace \u00b7 enterprise RAG", "Web-search API $10/1k", "Inline, easy-to-check", "Research in work context" ], [ "Elicit", "$0\u2013169/mo", "Usage tiers per plan", "138M papers + 545k trials", "Your PDFs / library", "Yes (Pro+)", "Per-claim to papers", "Systematic reviews" ], [ "Consensus", "Free + premium", "Deep search mode", "Peer-reviewed literature", "No", "No", "Evidence-meter per question", "Quick evidence checks" ], [ "Exa", "Usage: $7/1k search \u00b7 $12\u201315/1k deep", "API-metered", "Own index: 80B docs, 1.4T URLs", "Via your stack", "Core product", "Structured, grounded", "Research inside your product" ], [ "Kagi", "$5\u201325/mo", "Research mode on Ultimate $25", "Own + partner indexes", "No", "Limited", "Assistant cites sources", "Ad-free, de-SEO'd search" ] ] }, "rules": [ { "if": "You ask dozens of sourced questions a day and want the fastest answer-with-citations loop", "then": "Perplexity Pro at $20 \u2014 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", "then": "ChatGPT deep research \u2014 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", "then": "Gemini Deep Research with Workspace sources \u2014 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", "then": "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", "then": "Buy the API layer, not a subscription: Exa for index quality, Parallel for price-performance ($1/1k entry), Tavily for agent-framework ubiquity." } ], "field": [ { "name": "Exa", "maker": "Exa Labs", "note": "The research engine for agents: own 80B-doc/1.4T-URL index, search + deep-research APIs; $85M Series B at $700M (Sept 2025), ~$10M revenue at the time \u2014 powers Notion Research Mode, Cursor, Devin", "url": "https://exa.ai", "oss": false, "entry": "free credits, then $7/1k searches", "status": "active" }, { "name": "Consensus", "maker": "Consensus NLP", "note": "Academic search with per-question evidence meter over peer-reviewed literature; deep search mode. Exact pricing unverifiable on its JS-rendered pricing page \u2014 free tier + paid premium", "url": "https://consensus.app", "oss": false, "entry": "free tier", "status": "active" }, { "name": "You.com / ARI", "maker": "You.com", "note": "Consumer answer engine pivoted to enterprise research APIs; ARI deep-research agent (Feb 2025, 400+ sources per report) now sold as Research/Finance APIs; $100M Series C at $1.5B (Sept 2025), customers incl. OpenAI, NVIDIA", "url": "https://you.com", "oss": false, "entry": "API, custom", "status": "active" }, { "name": "Kagi", "maker": "Kagi Inc.", "note": "Paid, ad-free, de-SEO'd search with Assistant research mode on Ultimate; 73,842 paying members, ~1.06M queries/day (Aug 6, 2026) \u2014 beloved, resolutely niche", "url": "https://kagi.com", "oss": false, "entry": "$5\u201325/mo", "status": "active" }, { "name": "Undermind", "maker": "Undermind Inc.", "note": "Deep scientific-literature search reading full texts and figures; claims 10x better recall than Google Scholar; 1,000+ GSK scientists using it (2026)", "url": "https://undermind.ai", "oss": false, "entry": "free \u00b7 Pro $16/mo", "status": "active" }, { "name": "SciSpace", "maker": "PubGenius Inc.", "note": "Literature review + chat-with-PDF over a claimed 280M+ papers, 9.6M+ researchers claimed; broad but shallower rigor than Elicit", "url": "https://scispace.com", "oss": false, "entry": "free tier", "status": "active" }, { "name": "scite", "maker": "Research Solutions", "note": "Smart Citations \u2014 classifies citations as supporting/contrasting; 370k+ users; the citation-context layer other tools lack", "url": "https://scite.ai", "oss": false, "entry": "subscription", "status": "active" }, { "name": "Liner", "maker": "Liner (Aurumplanet)", "note": "Korean academic search/answer engine with credit model and reasoning modes; consumer-scale but stats unverified this cycle", "url": "https://getliner.com", "oss": false, "entry": "free tier", "status": "active" }, { "name": "M365 Copilot Researcher", "maker": "Microsoft", "note": "OpenAI deep-research model + Copilot orchestration over work email, meetings, files, and web, with Salesforce/ServiceNow/Confluence connectors; rolled out from Apr 2025 to Copilot-licensed orgs \u2014 the enterprise-procurement path", "url": "https://www.microsoft.com/en-us/microsoft-365/blog/2025/03/25/introducing-researcher-and-analyst-in-microsoft-365-copilot/", "oss": false, "entry": "M365 Copilot license (~$30/seat)", "status": "active" }, { "name": "Grok research", "maker": "xAI", "note": "DeepSearch/DeeperSearch branding (Feb 2025) has faded from x.ai \u2014 research now sold as parallel-agent reasoning with live web + X integration in SuperGrok; strongest on breaking/X-native topics", "url": "https://x.ai/grok", "oss": false, "entry": "free \u00b7 SuperGrok sub", "status": "active" }, { "name": "Tavily", "maker": "Tavily (Nebius subsidiary)", "note": "Search/extract/crawl API for agents \u2014 2M+ developers, 300M+ requests/mo; raised $25M Series A then became a Nebius subsidiary (2026)", "url": "https://tavily.com", "oss": false, "entry": "free tier + usage", "status": "acquired" }, { "name": "Parallel Web Systems", "maker": "Parallel (Parag Agrawal)", "note": "Search/Task/FindAll APIs with calibrated-confidence citations; claims 91% SimpleQA at $8 CPM vs Exa 89.3%/$20 and Tavily 72%/$23 \u2014 vendor benchmark, price-war entry at $1/1k", "url": "https://parallel.ai", "oss": false, "entry": "$1/1k requests + credits", "status": "active" }, { "name": "STORM / Co-STORM", "maker": "Stanford OVAL", "note": "Open-source Wikipedia-style report generator that seeded the deep-research pattern; live demo still up, research-grade UX", "url": "https://storm.genie.stanford.edu", "oss": true, "entry": "free", "status": "active" }, { "name": "NotebookLM", "maker": "Google", "note": "Source-grounded notebook, not a web research engine \u2014 belongs with the knowledge substrate (see Kw), but its Deep Research source-gathering blurs the line", "url": "https://notebooklm.google.com", "oss": false, "entry": "free \u00b7 AI Pro", "status": "active" }, { "name": "Google AI Mode", "maker": "Google", "note": "Cited AI answers free inside Search proper (Mar 2025, then global rollout) \u2014 not a product you buy, but the gravity well compressing every standalone answer engine's free tier", "url": "https://google.com/aimode", "oss": false, "entry": "free", "status": "active" }, { "name": "Brave Search API", "maker": "Brave", "note": "Independent-index search API used for LLM grounding; the privacy-first alternative in the agent-search layer", "url": "https://brave.com/search/api/", "oss": false, "entry": "free tier + usage", "status": "active" }, { "name": "Phind", "maker": "Phind Inc.", "note": "Developer answer engine of the 2023\u201324 wave; root page returned 404 to our crawler (Aug 2026) and momentum has visibly stalled \u2014 verify before depending on it", "url": "https://www.phind.com", "oss": false, "entry": "unverified", "status": "fading" }, { "name": "Neeva", "maker": "Neeva", "note": "The original paid ad-free search bet \u2014 shut down consumer product 2023, team acquired by Snowflake; Kagi inherited the thesis", "url": "https://neeva.com", "oss": false, "entry": "\u2014", "status": "dead" }, { "name": "Elicit (Ought era)", "maker": "Ought", "note": "Nonprofit research lab that built Elicit; spun the product out as a PBC in 2023 \u2014 listed so the lineage isn't mistaken for a separate tool", "url": "https://ought.org", "oss": false, "entry": "\u2014", "status": "renamed" } ], "signals": [ { "fact": "Perplexity: $232M revenue 2025 \u2192 $500M ARR by Apr 2026 (+335% YoY), $21.2B Series E-6 valuation (early 2026), subscription-first pivot Feb 2026", "src": "https://sacra.com/c/perplexity/" }, { "fact": "Deep research went from novelty (OpenAI, Feb 2025) to table stakes in ~12 months \u2014 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": "https://en.wikipedia.org/wiki/Deep_research" }, { "fact": "The agent-search API layer consolidated in Sept 2025\u20132026: 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": "https://sacra.com/c/exa/" }, { "fact": "Google AI Mode (Mar 2025 launch, global rollout through 2025) gives away cited AI answers inside Search \u2014 the structural squeeze on every standalone answer engine's free tier", "src": "https://en.wikipedia.org/wiki/AI_Mode" }, { "fact": "Paid human search stays viable but niche: Kagi at 73,842 paying members and ~1.06M queries/day (Aug 6, 2026)", "src": "https://kagi.com/stats" }, { "fact": "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": "https://elicit.com" } ], "notes": "Ranking criteria: speed-to-trustworthy-cited-answer, depth of autonomous research, grounding in the user's own data, and evidence discipline of the output \u2014 editorial picks, no affiliate consideration. Conflicts resolved: Perplexity's valuation \u2014 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 \u2014 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 \u2014 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 \u2192 Kw \u00b7 Knowledge Wiki; vector search infra \u2192 Vd; meeting transcripts \u2192 Mt; agent memory \u2192 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.", "sources": [ "https://en.wikipedia.org/wiki/Perplexity_AI", "https://en.wikipedia.org/wiki/Comet_(web_browser)", "https://sacra.com/c/perplexity/", "https://openai.com/index/introducing-deep-research/", "https://en.wikipedia.org/wiki/Deep_research", "https://help.openai.com/en/articles/10500283-deep-research-faq", "https://gemini.google/overview/deep-research/", "https://support.google.com/gemini/answer/15719111", "https://one.google.com/about/google-ai-plans/", "https://claude.com/blog/research", "https://support.claude.com/en/articles/11088861-using-research-on-claude", "https://claude.com/pricing", "https://elicit.com", "https://elicit.com/pricing", "https://exa.ai", "https://exa.ai/pricing", "https://sacra.com/c/exa/", "https://kagi.com/pricing", "https://kagi.com/stats", "https://you.com", "https://en.wikipedia.org/wiki/You.com", "https://undermind.ai", "https://tavily.com", "https://parallel.ai", "https://scispace.com", "https://scite.ai", "https://consensus.app", "https://www.microsoft.com/en-us/microsoft-365/blog/2025/03/25/introducing-researcher-and-analyst-in-microsoft-365-copilot/", "https://en.wikipedia.org/wiki/AI_Mode", "https://x.ai/grok", "https://docs.perplexity.ai" ], "element": { "number": 18, "name": "Research Engine", "group": "Knowledge & Memory", "essential": true, "edition": "v2026.Q3", "revision": "r7", "license": "CC BY 4.0 \u2014 cite elems.ai", "url": "https://elems.ai/e/re.html" } }