"Deep research" denotes AI tools that go beyond returning search results or a single chat answer: autonomous agents that plan a multi-step strategy, execute iterative searches, read many sources, reconcile contradictions, and deliver a structured citation-backed report.
Deep research definition and approachopenai.com
The category crystallized in late 2024–early 2025 when OpenAI, Google, Perplexity, and xAI each launched such agents.
OpenAI Deep Research launchopenai.com
Google Gemini Deep Research launchblog.google
xAI Grok 3 launchx.ai
Alongside sit older science-focused tools (Elicit, Consensus, Undermind) and premium enterprise intelligence platforms (AlphaSense).
Elicit scientific literature toolelicit.com
Consensus scientific search engineconsensus.app
AlphaSense financial intelligence platformwww.alpha-sense.com
Six dimensions were evaluated: capabilities & methodology; output quality/depth; use cases/audience; pricing/access; strengths/weaknesses; and differentiator. Ranking weights demonstrated citable capability and breadth of applicability.
Agentic feature in ChatGPT that autonomously browses, plans multi-step research, reads dozens of sources, and produces a long-form cited report, using a fine-tuned version of o3.
OpenAI Deep Research capabilities and o3 modelopenai.com
Reports reportedly run ~5–30 minutes per query with substantial long-form output and numerous inline citations (exact lengths vary by task). OpenAI reported 26.6% on HLE at launch vs ~3.3% for GPT-4o, and reported strong GAIA performance.
OpenAI Deep Research benchmark results on HLE and GAIA, and run timeopenai.com
Released February 2, 2025 to the ChatGPT Pro tier.
OpenAI Deep Research release date (Pro tier)openai.com
Pricing: ChatGPT Plus $20/mo; ChatGPT Pro $200/mo.
OpenAI ChatGPT subscription pricing tiersopenai.com
Audience: research professionals, analysts, knowledge workers, academics, consultants. Strengths: highest published benchmark at launch; strong synthesis. Weaknesses: slow; can hallucinate citations; Pro price high. Differentiator: first major lab to ship an autonomous deep-research agent.
Multi-step agent in Gemini Advanced; browses the Google index; users review/edit the research plan before execution; exports reports to Google Docs.
Google Gemini Deep Research features and capabilitiesblog.google
Reportedly powered by Gemini 1.5 Pro at launch. Launched December 11, 2024 for Gemini Advanced subscribers.
Google Gemini Deep Research launch date and launch modelblog.google
Pricing: Google One AI Premium $19.99/mo.
Google One pricing plansone.google.com
Audience: consumers, students, business professionals, Workspace users. Strengths: broad Google index coverage; editable plan; Docs export; competitive price. Weaknesses: plan-approval step adds friction; no independently published benchmarks. Differentiator: research-plan transparency/editability before execution.
Deep-research mode performing many iterative web searches, reading full pages, synthesizing a cited report with numbered inline citations.
Perplexity Deep Research capabilitieswww.perplexity.ai
Launched approximately February 2025. Pricing: Perplexity Pro $20/mo (or $200/yr).
Perplexity Pro pricingwww.perplexity.ai
The free tier offers limited Deep Research access, while Pro provides a substantially higher daily allowance. Audience: journalists, analysts, researchers, knowledge workers. Strengths: competitive price; relatively fast; strong inline citations; real-time web. Weaknesses: output structure less systematic than OpenAI's; hallucination risk. Differentiator: speed and affordability.
Microsoft 365 Copilot offers agentic, multi-step capabilities integrated with Word, Excel, Teams, and SharePoint. At Microsoft Ignite on November 19, 2024, Microsoft announced AI agents for Microsoft 365 Copilot designed to automate and execute multi-step business processes.
Microsoft Ignite (Nov 19, 2024): AI agents for Microsoft 365 Copilotblogs.microsoft.com
Microsoft subsequently introduced a dedicated deep-research "Researcher" agent for Microsoft 365 Copilot that conducts multi-step research across both internal organizational data and the open web; it is reported to be built on an OpenAI deep-research reasoning model (specifics per Microsoft's later announcements).
Pricing: Microsoft 365 Copilot is licensed at approximately $30/user/month on top of a Microsoft 365 subscription.
Microsoft 365 Copilot pricingwww.microsoft.com
Audience: enterprise knowledge workers, business analysts, consultants in Microsoft-centric orgs. Strengths: deep M365 integration; internal-and-external data fusion; advanced reasoning-model quality. Weaknesses: expensive; not for consumers; limited early rollout. Differentiator: simultaneous internal-and-external data fusion within M365 workflows.
DeepSearch mode in Grok 3 that iteratively searches the web and X; real-time info; visible "thinking" mode; powered by Grok 3.
Grok 3 DeepSearch capabilitiesx.ai
Launched February 17, 2025. Grok 3 and DeepSearch are offered through X's paid Premium tiers and via the standalone Grok app and site (subscription pricing varies).
Grok 3 launch date and availabilityx.ai
X Premium subscription tiershelp.x.com
At launch, xAI reported strong performance on AIME 2025 and GPQA.
Grok 3 benchmark performance on AIME 2025 and GPQAx.ai
Audience: X power users, journalists, analysts monitoring social/news in real time. Strengths: real-time X/social-web data; visible reasoning; strong reported reasoning benchmarks. Weaknesses: X noise/misinformation risk; less structured output; coupled to X ecosystem. Differentiator: real-time integration of X social data alongside open-web research.
AI research assistant for scientific literature review, evidence synthesis, and structured extraction; searches a large corpus of academic literature built on Semantic Scholar; extracts population/intervention/outcome/effect-size fields for systematic reviews.
Elicit scientific research tool capabilitieselicit.com
Developed by Ought (an AI safety–focused nonprofit), operational since ~2021, with commercial tiers introduced 2023–2024.
Elicit background and developmentelicit.com
Pricing: a free tier plus a paid Elicit Plus plan; see Elicit's pricing page for current rates.
Elicit pricing pageelicit.com
Audience: academic researchers, systematic reviewers, clinical researchers, grad students. Strengths: low citation-hallucination (database-grounded); structured extraction; systematic-review support. Weaknesses: scientific literature only; narrower general coverage. Differentiator: structured PICO-style evidence extraction with database-grounded citations.
AI scientific search engine surfacing peer-reviewed evidence and distilling expert agreement; indexes a large corpus via Semantic Scholar; "Consensus Meter" visualizes agreement; GPT-4-powered synthesis.
Consensus scientific search engine capabilitiesconsensus.app
Beta 2022; public launch with GPT-4 in 2023.
Consensus launch historyconsensus.app
Pricing: a free tier plus a paid Premium plan; see Consensus's pricing page for current rates.
Consensus pricing pageconsensus.app
Audience: students, journalists, policymakers, science communicators. Strengths: fast consensus-gauging; consumer-friendly; accurate database-grounded citations; affordable. Weaknesses: academic literature only; can oversimplify contested debates. Differentiator: Consensus Meter.
Deep-research tool for academic science; recursive literature search from seed papers across related work/citations/co-citations via Semantic Scholar; returns ranked, relevance-explained paper lists.
Undermind academic research tool capabilitieswww.undermind.ai
Launched approximately 2023–2024. Pricing: free tier plus a paid plan (verify at live site). Audience: academic researchers in specialized subfields, PhD students. Strengths: thorough and recursive; surfaces non-obvious papers; explains relevance. Weaknesses: academic literature only; ranked lists rather than narrative synthesis. Differentiator: recursive citation-graph traversal.
AI search engine with a dedicated "Research" mode performing multi-step searches and cited synthesis; privacy-first positioning.
You.com research capabilitiesyou.com
Launched November 2021; dedicated research mode added ~2023–2024. Pricing: free tier plus a paid YouPro plan (verify current rates at the live site). Audience: privacy-conscious researchers, students, general users. Strengths: privacy-first; free tier; accessible. Weaknesses: less analytically deep than top tier; lower brand recognition. Differentiator: privacy-first architecture.
AI market-intelligence platform for finance/investment/corporate strategy; indexes a reported 300M+ documents incl. SEC filings, earnings transcripts, broker research, premium news, trade pubs, and expert interviews (via Tegus); semantic search ("Smart Synonyms"), summarization, sentiment analysis, real-time alerts, generative features.
AlphaSense financial intelligence platform capabilitieswww.alpha-sense.com
Founded 2011; significant AI rollout 2023–2025. Pricing: enterprise-only, no public list; third-party estimates ~$20,000–$40,000+/yr/user (third-party estimate, not vendor-published). Audience: investment analysts, equity researchers, corporate strategy, M&A, consultants. Strengths: premium curated/licensed corpus; best-in-class financial intelligence; semantic search; Tegus transcripts. Weaknesses: very expensive; enterprise-only; not general-purpose. Differentiator: premium licensed financial content + purpose-built financial AI workflows.
Pricing table (approximate, as of early 2025; verify at each provider's live site):
| Service | Pricing |
|---|---|
| OpenAI Deep Research | $20/mo (Plus) / $200/mo (Pro) |
| Gemini Deep Research | $19.99/mo (Google One AI Premium) |
| Perplexity Deep Research | $20/mo or $200/yr (Pro) |
| Microsoft Copilot Researcher | ~$30/user/mo (M365 Copilot license) |
| Grok DeepSearch | Via X Premium tiers / standalone Grok app (price varies) |
| Elicit | Free + paid Elicit Plus — see pricing page |
| Consensus | Free + paid Premium — see pricing page |
| Undermind | Free + paid tier — verify at site |
| You.com | Free + paid YouPro — verify at site |
| AlphaSense | Enterprise-only; ~$20K–40K+/yr/user (third-party estimate) |
OpenAI ChatGPT pricingopenai.com
Google One pricing plansone.google.com
Perplexity Pro pricingwww.perplexity.ai
Microsoft 365 Copilot pricingwww.microsoft.com
Benchmark results:
| Service | Benchmark Results |
|---|---|
| OpenAI Deep Research | 26.6% HLE at launch; ~3.3% GPT-4o baseline; strong GAIA performance (per OpenAI) |
| Grok 3 DeepSearch | Reported strong AIME 2025 and GPQA performance at launch (per xAI) |
| All others | No independently published benchmark results |
OpenAI Deep Research benchmark resultsopenai.com
Grok 3 benchmark performancex.ai
Source characteristics:
| Source Type | Services |
|---|---|
| Open web | OpenAI, Gemini, Perplexity, You.com |
| Web + X social data | Grok DeepSearch |
| Web + internal M365 data | Microsoft Copilot Researcher |
| Academic/scientific literature only | Elicit, Consensus, Undermind |
| Premium curated financial/business content | AlphaSense |
Humanity's Last Exam (HLE) benchmark results (approximate, as of early 2025; note that only OpenAI has a vendor-published HLE score in this comparison):
Selected vendor-published consumer/prosumer monthly subscription prices (USD, approximate, as of early 2025). Tools without vendor-published monthly pricing — including AlphaSense (enterprise contracts) and services whose current rates should be confirmed on their pricing pages — are omitted from the chart:
Because no capabilities, pricing, features, or benchmarks could be verified, none are asserted here; asserting unverified specifics would constitute fabrication. Possible explanations include a post-research-cutoff launch, a small/niche offering not yet widely covered, a pre-launch or parked domain, a near-miss name for another service, or non-existence as an active product. Deeperer.com is treated as an unverified and potentially emerging entrant and is not ranked.
To enter a future ranking, it would need: documented methodology, evidence of output depth/quality, transparent pricing/access terms, and ideally independently reproducible benchmark results. Readers should verify the current live site and corroborate any claims independently.
Choose by use case:
OpenAI Deep Research capabilities and benchmarksopenai.com
Perplexity Pro pricing and capabilitieswww.perplexity.ai
Google Gemini Deep Research featuresblog.google
Microsoft 365 Copilot AI agents announced at Igniteblogs.microsoft.com
Grok 3 DeepSearch capabilitiesx.ai
Elicit scientific research toolelicit.com
Consensus scientific search engineconsensus.app
Undermind academic research toolwww.undermind.ai
AlphaSense financial intelligence platformwww.alpha-sense.com
The category moved from essentially nonexistent to crowded within ~two months (December 2024–February 2025). Future competition is likely to concentrate on citation reliability, source breadth and licensing, enterprise workflow integration, and price-to-depth tradeoffs.
Google Gemini Deep Research launch December 11, 2024blog.google
OpenAI Deep Research launch February 2025openai.com
Grok 3 launch February 17, 2025x.ai
OpenAI described Deep Research as an agentic capability that conducts multi-step research on the internet for complex, lengthy tasks and produces a thorough, fully documented report at a quality and accuracy level approaching that of a research analyst.
OpenAI Deep Research descriptionopenai.com
Google described Gemini Deep Research as a tool that helps users quickly develop a research plan and conducts complex research for them, saving hours of work.
Google Gemini Deep Research descriptionblog.google
In its Grok 3 launch post, xAI described DeepSearch as a mode that iteratively searches the web and X to reason through complex questions and return well-sourced answers.
xAI Grok 3 DeepSearch descriptionx.ai
At Microsoft Ignite on November 19, 2024, Microsoft announced AI agents for Microsoft 365 Copilot built to automate and orchestrate multi-step business processes.
Microsoft Ignite (Nov 19, 2024) AI agents announcementblogs.microsoft.com
Produced by AI. A cluster of agents wrote this report from the cited sources. Check it against those sources before you act on it.