February 2, 2025ReleaseIntroducing deep researchAn agent that uses reasoning to synthesize large amounts ofonline information and complete multi-step research tasksfor you. Available to Pro users today, Plus and Team next.Try on ChatGPT00:00|01:01Listen to article8:54Share09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/1/36
Today we’re launching deep research in ChatGPT, a new agentic capability thatconducts multi-step research on the internet for complex tasks. It accomplishes intens of minutes what would take a human many hours.Deep research is OpenAI's next agent that can do work for you independently—yougive it a prompt, and ChatGPT will find, analyze, and synthesize hundreds of onlinesources to create a comprehensive report at the level of a research analyst. Poweredby a version of the upcoming OpenAI o3 model that’s optimized for web browsing anddata analysis, it leverages reasoning to search, interpret, and analyze massive amountsof text, images, and PDFs on the internet, pivoting as needed in reaction to informationit encounters.The ability to synthesize knowledge is a prerequisite for creating new knowledge. Forthis reason, deep research marks a significant step toward our broader goal ofdeveloping AGI, which we have long envisioned as capable of producing novelscientific research.Why we built deep researchDeep research is built for people who do intensive knowledge work in areas likefinance, science, policy, and engineering and need thorough, precise, and reliableresearch. It can be equally useful for discerning shoppers looking for hyper-personalized recommendations on purchases that typically require careful research,like cars, appliances, and furniture. Every output is fully documented, with clearcitations and a summary of its thinking, making it easy to reference and verify theinformation. It is particularly effective at finding niche, non-intuitive information thatwould require browsing numerous websites. Deep research frees up valuable time byallowing you to offload and expedite complex, time-intensive web research with justone query.Deep research independently discovers, reasons about, and consolidates insights fromacross the web. To accomplish this, it was trained on real-world tasks requiringbrowser and Python tool use, using the same reinforcement learning methods behindOpenAI o1, our first reasoning model. While o1 demonstrates impressive capabilities incoding, math, and other technical domains, many real-world challenges demand09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/2/36
extensive context and information gathering from diverse online sources. Deepresearch builds on these reasoning capabilities to bridge that gap, allowing it to takeon the types of problems people face in work and everyday life.How to use deep researchIn ChatGPT, select ‘deep research’ in the message composer and enter your query. TellChatGPT what you need—whether it’s a competitive analysis on streaming platformsor a personalized report on the best commuter bike. You can attach files orspreadsheets to add context to your question. Once it starts running, a sidebarappears with a summary of the steps taken and sources used.Deep research may take anywhere from 5 to 30 minutes to complete its work, takingthe time needed to dive deep into the web. In the meantime, you can step away orwork on other tasks—you’ll get a notification once the research is complete. The finaloutput arrives as a report within the chat – in the next few weeks, we will also beadding embedded images, data visualizations, and other analytic outputs in thesereports for additional clarity and context.Compared to deep research, GPT-4o is ideal for real-time, multimodal conversations.For multi-faceted, domain-specific inquiries where depth and detail are critical, deepresearch’s ability to conduct extensive exploration and cite each claim is the differencebetween a quick summary and a well-documented, verified answer that can be usableas a work product.GPT-4oHelp me find iOS and android adoption rates, % who want to learnanother language, and change in mobile penetration, over the past 10years, for top 10 developed and top 10 developing countries by GDP.Lay this info out in a table and separate stats into columns, and include09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/3/36
Deep research responds to the prompt in a highly detailed manner, providing side-by-side count-speciﬁc data for both top 10 developed and top 10 developing countes for easy reference andcompason. It uses that infoation to oer detailed maet-ent recommendations that areinfoed and usable.recommendations on markets to target for a new iOS translation appfrom ChatGPT, focusing on markets ChatGPT is currently active in.Compiling comprehensive data on iOS and Android adoption rates, thepercentage of populations interested in learning another language, andchanges in mobile penetration over the past decade for the top 10developed and developing countries by GDP is challenging due to datalimitations. However, based on available information, we can provideinsights and recommendations for targeting markets with a new iOStranslation app powered by ChatGPT.Deep researchHelp me find iOS and android adoption rates, % who want to learnanother language, and change in mobile penetration, over the past 10years, for top 10 developed and top 10 developing countries by GDP.Lay this info out in a table and separate stats into columns, and includerecommendations on markets to target for a new iOS translation appfrom ChatGPT, focusing on markets ChatGPT is currently active in.Top 10 Developed Economies (by GDP) – Smartphone OS Share,Language-Learning Interest, and Mobile Penetration GrowthCountryiOSMarketShareAndroidMarketShareInterest in NewLanguage(% of population)MobilePenetrationChange(2013→2023)09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/4/36
How it worksDeep research was trained using end-to-end reinforcement learning on hard browsingand reasoning tasks across a range of domains. Through that training, it learned toplan and execute a multi-step trajectory to find the data it needs, backtracking andreacting to real-time information where necessary. The model is also able to browseover user uploaded files, plot and iterate on graphs using the python tool, embed bothgenerated graphs and images from websites in its responses, and cite specificsentences or passages from its sources. As a result of this training, it reaches newhighs on a number of public evaluations focused on real-world problems.Humanity's Last ExamOn Humanity’s Last Exam , a recently released evaluation that tests AI across a broadrange of subjects on expert-level questions, the model powering deep research scoresa new high at 26.6% accuracy. This test consists of over 3,000 multiple choice andshort answer questions across more than 100 subjects from linguistics to rocketscience, classics to ecology. Compared to OpenAI o1, the largest gains appeared inchemistry, humanities and social sciences, and mathematics. The model poweringdeep research showcased a human-like approach by effectively seeking outspecialized information when necessary.ModelAccuracy(%)GPT-4o3.3Grok-23.8Claude 3.5 Sonnet4.3Gemini Thinking6.2OpenAI o19.1DeepSeek-R1*9.4OpenAI o3-mini(medium)*10.509/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/5/36
ModelAccuracy(%)OpenAI o3-mini (high)*13.0OpenAI deep rearch**26.6* Model is not multi-modal,evaluated on text-only subset.**with browsing + python toolsGAIAOn GAIA , a public benchmark that evaluates AI on real-world questions, the modelpowering deep research reaches a new state of the art (SOTA), topping the externalleaderboard . Encompassing questions across three levels of difficulty, successfulcompletion of these tasks requires abilities including reasoning, multi-modal fluency,web browsing, and tool-use proficiency.GAIALevel1Level2Level3Avg.67.9267.4442.3163.64DeepResearch(pass@1)74.2969.0647.667.36DeepRearch(con)78.73.2158.0372.571PreviousSOTAGAIA task examples09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/6/36
In Nature journal's Scientific Reports conference proceedings from2012, in the article that did not mention plasmons or plasmonics, whatnano-compound is studied? Don't use the prefix nano in your answer ifthere is one.[The remaining Chain-of-Thought messages and the final answer arehidden to protect the benchmark's ground truth answer.]Clarifying the searchI’m piecing together the request for an article from 2012 in ScientificReports that details a nano-compound, excluding any related toplasmonics. Likely, this involves filtering specific topics or compounds, likesilver or graphene.Piecing together detailsI'm sub-searching "Scientific Reports 2012 nanoparticle," likely seekingnano-compounds. Interested in filtering out plasmon-related topics,possibly focusing on silver, graphene, or quantum dots.[search action]Navigating search filtersI’m digging into advanced search filters on nature.com and ScientificReports for 2012, trying to pinpoint articles on amorphous carbonnanoparticles in food caramels.Identifying conference proceedingsThe search results reveal a conference proceeding titled "2D quasiperiodicplasmonic crystals" from 2012, likely found under the "ConferenceProceedings" category in Scientific Reports.[browse action]Reflecting on potential issues09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/7/36
I'm considering the reasons behind a web page misconception, likely dueto an outdated agent type or inadvertent blockage.[browse action]Considering dynamic optionsI'm thinking about a nature site with dynamic content or a cached version.Alternatively, trying a search for specific phrases might help.[search action]Pinpointing key featuresOK, I'm identifying four conference proceedings from 2012 in Sci Rep.Notably, some mention "plasmonic" and should be excluded.[browse action]Piecing together insightsI’m gathering info on accessing cached versions or snippet expansions forspecific titles to ensure comprehensive analysis.[search action]Pulling together the detailsI’m noting the progression from the 2012 European Materials ResearchSociety Spring Meeting to the scientific report, focusing on conferenceproceedings. This approach helps establish context and relevance.[search action]Tracking journal cluesTrying to connect conference proceedings from late 2012 with "ScientificReports" journal, focusing on plasmonic 2D quasiperiodic materials. Thisapproach feels methodical and precise.[search action]Analyzing event detailsNoting that the search results [8] include a page likely summarizing theEuropean Materials Research Society 2012 Spring Meeting, including09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/8/36
25 symposia.[browse action]Piecing together researchI’m examining a likely collection page with 4 articles, consideringplasmonic and metamaterial topics, and identifying key references fromthe European Materials Research Society 2012 Spring Meeting.[browse action]Pinpointing the focusI’m focusing on the 2012 conference proceedings in "Scientific Reports"from E-MRS, likely involving topics like "2D quasiperiodic plasmoniccrystals" and "Layered plasmonic cloaks to tailor the optical scattering atthe nanoscale."[browse action]Evaluating library sourcesThinking about special issue E-MRS 2012 Sci rep invited paper Monticoneand metamaterials lab at News Archives – 2012.[browse action]Assessing accessibilityI'm gathering info on potential site access issues, consideringoutdated status or blockages. Also, evaluating the Monticone site for alist reference.[browse action]Piecing together E-MRS 2012I'm trying to link "Layered Plasmonic Cloaks" to E-MRS 2012. Consideringother possible topics like graphene and catalysis.[browse action]Identifying relevant topics09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/9/36
Expert-Level TasksIn an internal evaluation of expert-level tasks across a range of areas, deep researchwas rated by domain experts to have automated multiple hours of difficult,manual investigation. Pass Rate vs Max Tool Calls00.050.10.150.2Pass Rate020406080Max Tool CallsThe more the model browses and thinks about what its browsing,the better it does, which is why giving it time to think is impoant.Expert-level task examplesI’m gathering info on topics like graphene, photovoltaics, andnanostructured compounds, while excluding plasmonic-related ones. Thisleads me to focus on nanotechnology aspects more broadly.[browse action]Adjusting my focusI’m broadening the search parameters by entering specific site andscholar identifiers to pinpoint articles about 2D quasiperiodic plasmoniccrystals in Nature Scientific Reports from 2012.[search action]Identifying non-plasmonic possibilitiesI'm narrowing down the search to identify articles from the 2012 E-MRSspecial issue in Scientific Reports. This involves checking titles for non-plasmonic topics, like graphene or functional oxides. Progress is evidentthrough focused searches and exclusion of plasmonic articles.[search action]Piecing together cluesI'm analyzing interactions in a graphene superlattice, likely from a Sci Reparticle. Considering titles by authors like Padooru, Yakovlev, Chen, and Alu.[search action]Piecing together the puzzleI’m gathering info on E-MRS 2012 Spring Meeting's Sci Rep proceedings,focusing on invited papers, without plasmonic phenomena. Typesunder review are graphene, catalysis, energy materials, nanomedicine,and sensors.[search action]09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/10/36
Time saved on task: 4 hoursI am a researcher interested in bringing mixed-gas sorptioncapabilities to my lab. Please discuss the differences between pure-and mixed-gas sorption for glassy polymers, how the dual-modesorption model can be used to predict mixed-gas sorption behavior inglassy polymers (include equations where applicable), and whatchallenges there are in accurately predicting pure- and mixed-gassorption using the dual-mode sorption model. Only include open-source information as of October 1, 2024.Pure vs. mixed-gas sorption in glassy polymers: In a glassy polymer(below ), a single gas typically exhibits a concave sorption isotherm:uptake rises sublinearly with pressure, unlike the linear Henry’s-lawbehavior above . This is attributed to “dual-mode” sorption: somepenetrant dissolves in the dense polymer matrix (obeying Henry’s law)while additional penetrant saturates specific microvoid “sites” (aLangmuir-type adsorption in the polymer’s nonequilibrium freevolume). By contrast, when two or more gases are present, theycompete for those limited Langmuir sorption sites. As a result, eachcomponent’s sorbed concentration in a mixture is generally lowerthan it would be at the same partial pressure alone. In particular theless soluble (lower-affinity) gas tends to be more strongly suppressed,often causing the mixed-gas solubility selectivity (ratio ofconcentrations) to exceed the ideal value predicted from pure-gasisotherms. For example, in CO₂/CH₄ mixtures both components sorbless than in their pure-gas cases (especially CH₄), so the CO₂/CH₄sorption selectivity is higher under mixture conditions. In extremecases, mixed-gas effects can even reverse performance trendsobserved with single gases. (We note that unlike competitive sorption,synergistic sorption – where one gas increases the sorption of anotherT gT g09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/11/36
via polymer swelling – is usually minor in rigid high-free-volumepolymers under moderate pressures.)Dual-mode sorption model and mixed-gas prediction: The dual-modesorption (DMS) model quantitatively captures glassy polymer sorptionby superposing a Henry’s-law term and a Langmuir term. For a singlegas, the sorbed concentration  (e.g. in cm³(STP) gas per cm³polymer) is written as:Here  is the Henry’s law coefficient for dissolution,  is theLangmuir capacity constant (proportional to the polymer’s unrelaxedfree-volume or “hole” capacity), and  is the Langmuir affinityconstant. This empirical formula yields the characteristic concaveisotherm of glassy polymers and reduces to simple Henry’s law at theglass-transition (since  as ). The same threeparameters () are obtained by fitting pure-gas sorption dataand are then used to predict mixture sorption. For a binary mixture ofgases A and B (with partial pressures ), the DMS modelassumes each gas obeys its own Henry’s term and that both competefor the same Langmuir sites. One convenient form is to write the totalsorbed concentration as the sum of contributions from each gas:which implicitly means the concentration of component i in themixture is . This reduces to the single-gasformula when the other component’s pressure is zero. In practice,partial pressure in these equations is often replaced by fugacity foraccuracy at high pressures, though using fugacities or pressures givesnearly the same results if applied consistently. The DMS mixed-gasmodel predicts the competitive sorption effect observedexperimentally – namely, each gas’s uptake is depressed relative to itspure-gas sorption isotherm.Challenges and limitations: Although the dual-mode model is widelysuccessful in correlating sorption data, it has notable limitations inCC(p)=k p+D .1+bpC bpH′k DC H′bC →H′0T→T gk ,C ,bDH′p ,p ABC =totk p +D,AAk p +D,BB ,1+b p +b p AABBC b p +C b p H,A′AAH,B′BBC =ikp+D,ii 1+ b p ∑jjjC b p H,i′ii09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/12/36
predicting sorption behavior. First, the DMS model is empirical – itsfitted parameters are not universal constants but can vary withpolymer formation history, physical aging, and temperature/pressurerange. This means a parameter set obtained under one set ofconditions may lack predictive power outside its calibration range. Forinstance, sorption isotherms measured only over a limited pressurerange can be fit well by DMS, but extrapolating beyond that(especially approaching plasticization pressures) may be unreliable.Indeed, multiple distinct combinations of , ,  can sometimes fitthe same pure-gas isotherm almost equally well – yet they yieldmarkedly different mixed-gas predictions. This sensitivity underminesconfidence in mixture forecasts, as one study showed that usingdifferent pure-gas fit sets led to only qualitative agreement with somemixed-CO₂/CH₄ data.Furthermore, DMS inherently assumes a fixed population of Langmuirsites (microvoids) and ignores polymer relaxation or swelling inducedby sorbate. In reality, penetrant-induced swelling (plasticization) canincrease free volume and sorption capacity, especially for highlycondensable gases at high activity – an effect the basic DMS modelcannot capture. In binary sorption, this means any synergisticincrease in sorption due to a second gas (or due to polymer dilation) isunaccounted for in the standard model. Conversely, the model alsocannot describe sorption isotherm shapes that deviate from concaveLangmuir-type behavior (e.g. sigmoidal isotherms caused by specificpolymer-penetrant interactions or clustering). For example, glassypolymers absorbing polar molecules like alcohols can show sigmoidalsorption isotherms, which the simple dual-mode framework fails torepresent. Finally, accurately measuring mixed-gas sorption itself ischallenging, and discrepancies between idealized model assumptions(e.g. truly zero downstream pressure, no competitive diffusionlimitations) and experimental realities can add to prediction error.Despite these challenges, the dual-mode sorption model – withappropriate care – remains a valuable first approximation forinterpreting sorption in glassy polymers, and it highlights the key roleof competitive site occupancy in mixed-gas environments.Researchers often turn to more elaborate models (e.g. non-equilibriumk DC H′b09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/13/36
lattice fluid or thermodynamic theories) when the dual-mode model’sassumptions break down, but such models come at the cost ofadditional parameters or complexity. In summary, dual-mode sorptionprovides a useful baseline for pure- and mixed-gas sorption in glassypolymers, but one must be mindful of its empirical nature and theabove limitations when predicting multicomponent sorption behaviorin practice.Sources: Dual-mode sorption model and mixed-gas extension;competitive sorption effects in mixtures; model equations andparameters; limitations of DMS model; parameter sensitivity andpredictivity issues.Understanding sorption modelsI’m gathering info on pure and mixed-gas sorption in glassy polymers,focusing on the dual-mode sorption model and addressing challenges inmodeling predictions.[search action]Assessing open accessOK, let me see. The result from MDPI on mixed-gas sorption in glassypolymers is probably open access.[browse action][browse action]Examining sectionsI’m pulling together insights from the Dual Mode Sorption Model and itsextended analysis within the MDPI paper.[browse action]Clarifying key propertiesI’m examining the lab report to understand Henry's law constant,Langmuir affinity constant, and Langmuir capacity constant, which09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/14/36
describe gas sorption.[browse action]Piecing togetherI'm working through the possibility that "Equation (3)" and "Equation (4)"might be images or formula displays and considering consulting the PDFor searching for earlier references to understand them better.[browse action]Confirming title relevanceI’m verifying if the document from Freeman and Koros (1980) specificallymatches the reference "Model for sorption of mixed gases in glassypolymers" by checking its abstract and title.[search action]Examining article connectionsI noticed that the result from "pmc.ncbi.nlm.nih.gov" is likely thesame MDPI article, and I'm curious to see if it presents the text or asimplified version.[browse action][browse action][browse action]Gathering infoI'm gathering information about pure and mixed gas sorption in glassypolymers. This is a critical step in understanding the material's behaviorunder different gas environments.[search action]Comparing gas sorptionI noticed that carbon dioxide's presence in a mixture affects methanesorption in polymer, reducing methane's sorption compared to a puregas scenario.09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/15/36
[browse action]Linking conceptsI’m thinking through how the term "extended dual mode equation mixture"might relate to dual mode superconducting cable equations. This showspotential for connecting concepts across fields.[search action]Requesting review detailsI'm noting the request to review "Theories of Sorption and Transport inPolymer Membrane."[browse action]Checking PDF linkI’m curious about Result [2] and exploring the snippet, which involvesextending a dual-mode sorption model to a binary mixture and expressingconcentration of A in terms of partial pressures and constants.[search action]Predicting sorption behaviorThe dual-mode model anticipates that adding a second gas reducessorption at a constant partial pressure, unlike pure gas scenarios.[browse action]Fetching a PDFI'm gathering info on a nature.com PDF, possibly an older Yampol'skiarticle, but thinking about alternative sources like the University of Texasat Austin repository for a text snippet.[browse action]Evaluating repository optionsI’ve discovered that the UT repository result might not display correctly.Searching within it is uncertain. Alternatively, result [3] contains a snippet09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/16/36
describing the dual mode mixture sorption model, explaining competitiveeffects and its adequacy in modeling experimental data.[search action]Refining the focusI'm adjusting my approach to ensure the search targets the right context.[search action]Understanding sorption modelsI’m detailing the dual-mode sorption model for pure gases: Henrydissolving and Langmuir microvoid filling. For mixed gases, competitivesorption and non-idealities are emphasized.[search action]Exploring hero.epa.govThinking through hero.epa.gov's potential abstract or content.[browse action]Refining the modelI'm gathering insights on pure and mixed gas sorption, considering factorslike competition, plasticization, and parameter stability. Incorporatingmixed-gas data for more accurate predictions.[browse action]Addressing dual sorption model challengesParameter coupling and non-uniqueness impact predictions. Absence ofpolymer condition changes might lead to oversights.[browse action]Quantifying deviationI’m quantifying nonideal CO2-CH4 solubility in 6FDA-mPDA, showingmixed gas effects on solubility. Competitive sorption isn’t the sole causeof permeability selectivity decline.09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/17/36
Pass Rate on Expert-Level Tasks by Estimated Economic ValuePass Rate on Expert-Level Tasks by Estimated Hours0510152025Pass RateLowMediumHighVEstimated Economic Value0510152025Pass Rate1-3 hous)4-6 hours7-9 hours1Estimated Hours[browse action]Incorporating referencesI'm gathering citations, focusing on open-source materials like MDPI2019 and Vopicka 2014, to support the discussion on differencesand challenges in the dual-mode model for gas sorption inpolymeric membranes.[browse action]Proposing sectionsI'm organizing sections on pure vs mixed-gas sorption, exploring the dual-mode sorption model, and detailing challenges like parameter uncertaintyand mixed gas prediction.Organizing contentI’m focusing on the dual-mode sorption model, detailing pure and mixed-gas sorption behaviors, and highlighting key differences and challengeswith bullet points and structured sections for clarity.[search action]Choosing citationsOK, let me see. I’m leaning on [3], [20], [22], and [21] for their clear insightsinto dual-mode sorption and gas mixtures.Mapping out parametersI’m piecing together parameter names to avoid confusion. Sticking with(k), (C_H'), and (b) for clarity. The equations for pure gas and mixture aredetailed and clearly formatted.Addressing challengesI'm tackling multi-solution parameter fits, polymer plasticization, andtemporal effects. These intricacies make predicting pure and mixed gassorption tricky.Piecing togetherI’m structuring sections on sorption differences, the dual-mode model forpure gases, its extension to mixed gases, and the associated challenges.This progression seems logical and effective.09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/18/36
LimitationsDeep research unlocks significant new capabilities, but it’s still early and haslimitations. It can sometimes hallucinate facts in responses or make incorrectinferences, though at a notably lower rate than existing ChatGPT models, according tointernal evaluations. It may struggle with distinguishing authoritative information fromrumors, and currently shows weakness in confidence calibration, often failing toconvey uncertainty accurately. At launch, there may be minor formatting errors inreports and citations, and tasks may take longer to kick off. We expect all these issuesto quickly improve with more usage and time.AccessDeep research in ChatGPT is currently very compute intensive. The longer it takes toresearch a query, the more inference compute is required. We are starting with aversion optimized for Pro users today, with up to 100 queries per month. Plus andEstimatedeconomicvalueoftaskismorecoelatedwithpassratethan#ofhoursitwouldtakea00:00|00:0009/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/19/36
Team users will get access next, followed by Enterprise. We are still working onbringing access to users in the United Kingdom, Switzerland, and the EuropeanEconomic Area. All paid users will soon get significantly higher rate limits when we release a faster,more cost-effective version of deep research powered by a smaller model that stillprovides high quality results. In the coming weeks and months, we’ll be working on the technical infrastructure,closely monitoring the current release, and conducting even more rigorous testing.This aligns with our principle of iterative deployment. If all safety checks continue tomeet our release standards, we anticipate releasing deep research to Plus users inabout a month.What's nextDeep research is available today on ChatGPT web, and will be rolled out to mobile anddesktop apps within the month. Currently, deep research can access the open weband any uploaded files. In the future, you’ll be able to connect to more specialized datasources—expanding its access to subscription-based or internal resources—to makeits output even more robust and personalized.Looking further ahead, we envision agentic experiences coming together in ChatGPTfor asynchronous, real-world research and execution. The combination of deepresearch, which can perform asynchronous online investigation, and Operator, whichcan take real-world action, will enable ChatGPT to carry out increasingly sophisticatedtasks for you.Update on February 5, 2025: Deep research is now available to Pro users in the UnitedKingdom, Switzerland, and the European Economic Area.09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/20/36
Update on February 3, 2025: We conducted rigorous safety testing, preparednessevaluations, and governance reviews on the early version of o3 that powers deepresearch, identifying it as Medium  risk. We also ran additional safety testing to betterunderstand incremental risks associated with deep research's ability to browse theweb, and we have added new mitigations. We will continue to thoroughly test andclosely monitor the current limited release. We will share our safety insights andsafeguards for deep research in a system card when we widen access to Plus users.Footnotes1We found that the ground-truth answers for this dataset were widely leaked online and haveblocked several websites or URLs accordingly to ensure a fair evaluation of the model. AuthorsOpenAIResearch Leadstroction to DeeReseartroction to DeeResearPlay video09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/21/36
Isa Fulford, Zhiqing SunFoundational ContributorsAlex Tachard Passos, Alexandra Barr, Allison Tam, Charlotte Cole, Hyung Won Chung, Jason Wei,Jon Blackman, Scott Mayer McKinney, Valerie QiCore ContributorsResearchElaine Ya Le, Eric Mitchell, Eric Wallace, Hyung Won Chung, Ignasi Clavera, Leo Liu, Lorenz Kuhn,Louis Feuvrier, Max Schwarzer, Saachi Jain, Scottie Yan, Shunyu Yao, Vitchyr PongDeploymentCarpus Chang, Harry Zhao, Joseph Trasatti, Joshua Dickens, Matt Kaufer, Mike Trpcic, Minnia Feng,Neel Ajjarapu, Peter Vidani,  Sean FitzgeraldContributorsResearchAhmed El-Kishky, AJ Ostrow, Alexander Wei, Andrei Gheorghe, Andrew Kondrich, AndreyMishchenko, Anuj Nair, Behrooz Ghorbani, Brydon Eastman, Chak Li, Foivos Tsimpourlas, FrancisSong, Giambattista Parascandolo,Gildas Chabot, Hessam Bagherinezhad, Haitang Hu, HongyuRen, Henry Aspegren, Hunter Lightman, Ilya Kostrikov, Ilge Akkaya, James Lennon, Jean Harb,Jonathan Ward, Kai Chen, Katy Shi, Kevin Liu, Kevin Yu, Manuka Stratta, Marvin Zhang, MengyuanYan,  Mostafa Rohaninejad, Noam Brown, Phoebe Thacker, Raz Goan, Reah Miyara, Spencer Papay,Taylor Gordon, Wenda Zhou, Wenlei Xie, Yash Patil, Yann Dubois, Youlong Cheng, Yushi Wang,Wyatt Thompson+ all the contributors to o3.Safety SystemsAdam Kalai, Alex Beutel, Andrea Vallone, Andy Applebaum, David Robinson, Elizabeth Proehl, EvanMays, Grace Zhao, Irina Kofman, Jason Phang, Joaquin Quinonero Candela, Joel Parish, Kevin Liu,Kristen Ying, Lama Ahmad, Leon Maksin, Leyton Ho, Meghan Shah, Michele Wang, Miles Wang,Phillip Guo, Olivia Watkins, Owen Campbell-Moore, Patrick Chao, Sam Toizer, Samuel Miserendino,Sandhini Agarwal, Tejal Patwardhan, Tina Sriskandarajah, Troy Peterson, Yaodong Yu, Yunyun WangDeployment09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/22/36
Our ResearchResearcIexResearcOverviewResearcResideyLatestAdvaementsOpeI o1OpeI o1-miGPT-4oGPT-4o miraCtGPTExplore CtGPTTeamEnterpriseEducationPriciwoadrara OverviewFeaturesFor BusissOverviewCompanyAbout usOur CrterCareersBraMoreNewsoriesTerms & PoliciesTerms of UsePrivacy PolicycurityOtr PoliciesAdam Koppel, Adam Wells, Adele Li, Andy Applebaum, Andrey Malevich, Andrew Duberstein,Andrew Howell, Anton Tananaev, Ashley Tyra, Brandon Walkin, Bryan Ashley, Cary Bassin, CaryHudson, Cory Decareaux, Cristina Scheau, Derek Chen, Dibya Bhattacharjee, Drea Lopez, EricAntonow, Eric Burke, Filippo Raso, Fotis Chantzis, Freddie Sulit, Harris Cohen, Heather Whitney, JayDixit, Jeffrey Han, Jen Robinson, Jessica Shieh, Joel Parish, Kan Wu, Kevin Gladstone, KshitijWadhwa, Leo Vandriel, Leyton Ho, Liang Chen, Madeline Christian, Mamie Rheingold, Matt Jones,Michelle Fradin, Mike McClay, Mingxuan Wang, Nacho Soto, Niko Felix, Patrick Delaney, PaulMcMillan, Philip Pronin, Rodrigo Riaza Perez, Samuel Miserendino, Scott Ethersmith, StevenBaldwin, Thomas Dimson, Tomo Hiratsuka, Yaming Lin, Yara Khakbaz, Yining ChenLeadershipAkshay Nathan, Greg Brockman, Hannah Wong, Jakub Pachocki, Jerry Tworek, Johannes Heidecke,Josh Tobin, Liam Fedus, Mark Chen, Mia Glaese, Nick Turley, Sam Altman, Wojciech Zaremba09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/23/36
fetyfety Approachcurity & PrivacyPricira log inAPI PlatformPlatform OverviewPriciAPI log incumentationveloper ForumHelp CenterOpeI © 2012025Manage CookiesEnglishUnited States09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/24/36
09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/25/36
09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/26/36
09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/27/36
09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/28/36
09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/29/36
09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/30/36
09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/31/36
09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/32/36
09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/33/36
09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/34/36
09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/35/36
09/03/2025, 14:36Introducing deep research | OpenAIhttps://openai.com/index/introducing-deep-research/36/36