Introducing deep research Introducing deep research

_February 10, 2026 update:_ _You can now connect deep research to any MCP or app and restrict web searches to trusted sites, so you can focus on authenticated, industry-standard sources. You can also now track progress in real-time and interrupt to refine with follow-up prompts or new sources. We've updated the visual experience so it's easier to start, track, and review your research from end to end._

* * *

_July 17, 2025 update:_ _Deep research can now go even deeper and broader with access to a visual browser as part of ChatGPT agent. To access these updated capabilities, simply select “agent mode” from the dropdown in the composer and enter your query directly. The original deep research functionality remains available via the “deep research” option in the tools menu._

_April 24, 2025 update: We’re significantly increasing how often you can use deep research—Plus, Team, Enterprise, and Edu users now get 25 queries per month, Pro users get 250, and Free users get 5. This is made possible through a new lightweight version of deep research powered by a version of o4-mini, designed to be more cost-efficient while preserving high quality. Once you reach your limit for the full version, your queries will automatically switch to the lightweight version._

_February 25, 2025 update: All Plus users can now use deep research._

_February 5, 2025 update: Deep research is now available to Pro users in the United Kingdom, Switzerland, and the European Economic Area._

Today we’re launching deep research in ChatGPT, a new agentic capability that conducts multi-step research on the internet for complex tasks. It accomplishes in tens of minutes what would take a human many hours.

Deep research is OpenAI's next agent that can do work for you independently—you give it a prompt, and ChatGPT will find, analyze, and synthesize hundreds of online sources to create a comprehensive report at the level of a research analyst. Powered by a version of the upcoming OpenAI o3 model that’s optimized for web browsing and data analysis, it leverages reasoning to search, interpret, and analyze massive amounts of text, images, and PDFs on the internet, pivoting as needed in reaction to information it encounters.

The ability to synthesize knowledge is a prerequisite for creating new knowledge. For this reason, deep research marks a significant step toward our broader goal of developing AGI, which we have long envisioned as capable of producing novel scientific research.

Deep research is built for people who do intensive knowledge work in areas like finance, science, policy, and engineering and need thorough, precise, and reliable research. 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 clear citations and a summary of its thinking, making it easy to reference and verify the information. It is particularly effective at finding niche, non-intuitive information that would require browsing numerous websites. Deep research frees up valuable time by allowing you to offload and expedite complex, time-intensive web research with just one query.

Deep research independently discovers, reasons about, and consolidates insights from across the web. To accomplish this, it was trained on real-world tasks requiring browser and Python tool use, using the same reinforcement learning methods behind OpenAI o1, our first reasoning model. While o1 demonstrates impressive capabilities in coding, math, and other technical domains, many real-world challenges demand extensive context and information gathering from diverse online sources. Deep research builds on these reasoning capabilities to bridge that gap, allowing it to take on the types of problems people face in work and everyday life.

In ChatGPT, select ‘deep research’ in the message composer and enter your query. Tell ChatGPT what you need—whether it’s a competitive analysis on streaming platforms or a personalized report on the best commuter bike. You can attach files or spreadsheets to add context to your question. Once it starts running, a sidebar appears with a summary of the steps taken and sources used.

Deep research may take anywhere from 5 to 30 minutes to complete its work, taking the time needed to dive deep into the web. In the meantime, you can step away or work on other tasks—you’ll get a notification once the research is complete. The final output arrives as a report within the chat – in the next few weeks, we will also be adding embedded images, data visualizations, and other analytic outputs in these reports 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, deep research’s ability to conduct extensive exploration and cite each claim is the difference between a quick summary and a well-documented, verified answer that can be usable as a work product.

Deep research

Help me find iOS and android adoption rates, % who want to learn another language, and change in mobile penetration, over the past 10 years, for top 10 developed and top 10 developing countries by GDP. Lay this info out in a table and separate stats into columns, and include recommendations on markets to target for a new iOS translation app from ChatGPT, focusing on markets ChatGPT is currently active in.

Top 10 Developed Economies (by GDP) – Smartphone OS Share, Language-Learning Interest, and Mobile Penetration Growth

CountryiOS Market ShareAndroid Market ShareInterest in New Language

(% of population)Mobile Penetration Change (2013→2023)

United States

(USA)~58% iOS vs 42% Android _(see left)_High:

70% regret losing foreign-language ability (indicating strong desire to improve)56%→82%

smartphone penetration (↑26 pp)

Japan~69% iOS vs 31% Android _(see left)_High:

Over 50% want to learn English~30%→79%

(approx.) (↑~50 pp)

Germany~39% iOS vs 60% Android _(see left)_High:

86% of Europeans say everyone should speak ≥1 foreign language (strong pro-language attitude)40%→82%

(↑42 pp)

United Kingdom

(UK)~52% iOS vs 47% Android _(see left)_Moderate-High:

~73% have some interest (only 27% “no interest” in new languages)62%→82%

(↑20 pp)

France~35% iOS vs 64% Android _(see left)_High:

86% (EU average) favor multilingualism; English widely seen as important42%→83%

(↑40 pp)

Italy~30% iOS vs 69% Android _(see left)_High:

86% (EU avg.) favor language learning; 1 in 4 regret not learning another41%→85%

(↑44 pp)

Canada~60% iOS vs 40% Android _(see left)_Moderate:

Many bilingual (English/French); rising interest in third languages (no recent % data)56%→~80%

(↑ ~24 pp) (est., similar to US)

South Korea~24% iOS vs 76% Android _(see left)_Moderate:

Strong English education focus; ~40% of teens are learning English73%→~95%

(↑ ~22 pp) – already very high by 2013 (near saturation now)

Australia~55% iOS vs 45% Android _(see left)_Moderate:

~70% see learning languages as valuable (English dominant but interest in Asian languages growing)65%→~85%

(↑ ~20 pp) (est., high smartphone use)

Spain~20% iOS vs 79% Android _(see left)_High:

88% of Spaniards learned a foreign language in school (e.g. English); strong cultural interest55%→~85%

(↑ ~30 pp) (est., nearing Western Europe peers)

Top 10 Developing/Emerging Economies (by GDP) – _(ChatGPT availability considered)_

(notable stats)Mobile Penetration Change (2013→2023)

China

_^(ChatGPT restricted)_~24% iOS vs 75% Android _(see left)_Very High:

Hundreds of millions learning English (but ChatGPT not officially accessible)47%→68%

(↑21 pp)

India~4% iOS vs 95% Android _(see left)_High:

English compulsory in many schools; strong economic motivation (80% link English to higher income)~5%→46%

(↑ ~41 pp) (huge jump in smartphone adoption)

Brazil~18% iOS vs 82% Android _(see left)_High:

Great demand for English – only ~5% speak it, but most see it as key skill26%→67%

Russia

_^(ChatGPT limited)_~30% iOS vs 69% Android _(see left)_Moderate:

English learning common (compulsory in school); interest exists but ChatGPT availability is uncertain36%→74%

(↑38 pp)

Mexico~24% iOS vs 76% Android _(see left)_High:

Strong desire to learn English for US job/migration; ~50% of students study English in school (est.)37%→62%

(↑25 pp)

Indonesia~12% iOS vs 88% Android _(see left)_High:

English seen as gateway skill; government programs to improve English proficiency

(↑ ~40 pp)

Turkey~15% iOS vs 85% Android (est.)_(see left)_High:

English highly valued (universal in education);

80%

link English to better jobs30%→75%

(↑45 pp)

Saudi Arabia~24% iOS vs 76% Android _(see left)_Moderate:

English widely taught; 92% smartphone ownership (many already bilingual in English/Arabic)73%→~95%

(↑ ~22 pp) (early adopter; near saturation now)

South Africa~16% iOS vs 83% Android _(see left)_High:

Multilingual society; English and Afrikaans taught alongside native languages – cultural impetus to learn others40%→~60%

(↑ ~20 pp) (est., steady growth)

Argentina~12% iOS vs 88% Android _(see left)_High:

Strong interest in English (Argentina leads Latin America in English proficiency)31%→~70%

(↑ ~39 pp) (est., sharp rise with smartphone adoption)

Sources: Smartphone OS shares are from StatCounter/Statista data (2023–24). Language-interest metrics come from surveys by Preply, Eurobarometer, Babbel, etc. (e.g. UK 73% interest; US 70% regret lack of languages; Japan >50% want English; Pearson global survey). Mobile penetration is measured as smartphone ownership percentage – dramatic growth is noted in developing markets (e.g. Brazil +40 points) and high saturation in developed ones (e.g. UK ~82%) over the last decade.

Recommendations: Based on the above, prime targets for a new ChatGPT‑powered iOS translation app would be countries with large iOS user bases_and_strong language-learning demand, where ChatGPT is accessible. Notably, Japan, the U.S., UK, Canada, and Australia emerge as top choices. These markets pair high iOS adoption (50–60%+ share) with significant enthusiasm for language learning – for example, Japan’s iPhone dominance and widespread desire to learn English make it ideal. English-speaking countries like the US, UK, Canada, and Australia, despite being monolingual societies, show substantial interest in learning other languages (e.g. Spanish or French), and all have over 80% smartphone penetration and millions of iPhone users.

Among emerging economies, Mexico and Brazil stand out as secondary targets: although Android prevails, their sheer population sizes mean tens of millions of iPhone users exist, and interest in English is very high for career and education reasons. These Latin American markets (where ChatGPT is active) could yield strong adoption for an iOS-only app, especially given the motivation to learn English. Conversely, China and Russia are less feasible – China’s OS split skews Android and ChatGPT is not officially available there, and Russia faces access and regulatory uncertainties.

In summary, focus on iOS-heavy, pro-learning markets: Japan and Western countries (US, UK, Canada, Australia, plus high-income Europe like Germany and France to a lesser degree) should be the first-tier targets. They combine robust iOS usage with favorable language-learning attitudes. A second tier includes large-population developing markets (Mexico, Brazil, perhaps Saudi Arabia for its historically high iPhone uptake) where interest in English is strong – these can be approached once the app gains traction in the primary markets. Such a strategy ensures the app launches where it has the greatest alignment of platform (iOS) presence and user demand for translation/language tools, maximizing the likelihood of adoption and growth.

Deep research responds to the prompt in a highly detailed manner, providing side-by-side country-specific data for both top 10 developed and top 10 developing countries for easy reference and comparison. It uses that information to offer detailed market-entry recommendations that are informed and usable.

Deep research was trained using end-to-end reinforcement learning on hard browsing and reasoning tasks across a range of domains. Through that training, it learned to plan and execute a multi-step trajectory to find the data it needs, backtracking and reacting to real-time information where necessary. The model is also able to browse over user uploaded files, plot and iterate on graphs using the python tool, embed both generated graphs and images from websites in its responses, and cite specific sentences or passages from its sources. As a result of this training, it reaches new highs on a number of public evaluations focused on real-world problems.

On Humanity’s Last Exam⁠(opens in a new window), a recently released evaluation that tests AI across a broad range of subjects on expert-level questions, the model powering deep research scores a new high at 26.6% accuracy. This test consists of over 3,000 multiple choice and short answer questions across more than 100 subjects from linguistics to rocket science, classics to ecology. Compared to OpenAI o1, the largest gains appeared in chemistry, humanities and social sciences, and mathematics. The model powering deep research showcased a human-like approach by effectively seeking out specialized information when necessary.

| Model | Accuracy (%) |

| --- | --- |

| GPT-4o | 3.3 |

| Grok-2 | 3.8 |

| Claude 3.5 Sonnet | 4.3 |

| Gemini Thinking | 6.2 |

| OpenAI o1 | 9.1 |

| DeepSeek-R1* | 9.4 |

| OpenAI o3-mini (medium)* | 10.5 |

| OpenAI o3-mini (high)* | 13.0 |

| OpenAI deep research | 26.6 |

| Claude 3.5 Sonnet | 4.3 |

with browsing + python tools

On GAIA⁠(opens in a new window)1, a public benchmark that evaluates AI on real-world questions, the model powering deep research reaches a new state of the art (SOTA), topping the external leaderboard⁠(opens in a new window). Encompassing questions across three levels of difficulty, successful completion of these tasks requires abilities including reasoning, multi-modal fluency, web browsing, and tool-use proficiency.

| GAIA |

| --- |

| | Level 1 | Level 2 | Level 3 | Avg. |

| Previous SOTA(opens in a new window) | 67.92 | 67.44 | 42.31 | 63.64 |

| Deep Research (pass@1) | 74.29 | 69.06 | 47.6 | 67.36 |

| Deep Research (cons@64) | 78.66 | 73.21 | 58.03 | 72.57 |

In an internal evaluation of expert-level tasks across a range of areas, deep research was rated by domain experts to have automated multiple hours of difficult, manual investigation.

Pass Rate vs Max Tool Calls

The more the model browses and thinks about what its browsing, the better it does, which is why giving it time to think is important.

Pass Rate on Expert-Level Tasks by Estimated Economic Value

Pass Rate on Expert-Level Tasks by Estimated Hours

Estimated economic value of task is more correlated with pass rate than # of hours it would take a human – the things that models find difficult are different to what humans find time-consuming.

Deep research unlocks significant new capabilities, but it’s still early and has limitations. It can sometimes hallucinate facts in responses or make incorrect inferences, though at a notably lower rate than existing ChatGPT models, according to internal evaluations. It may struggle with distinguishing authoritative information from rumors, and currently shows weakness in confidence calibration, often failing to convey uncertainty accurately. At launch, there may be minor formatting errors in reports and citations, and tasks may take longer to kick off. We expect all these issues to quickly improve with more usage and time.

Deep research in ChatGPT is currently very compute intensive. The longer it takes to research a query, the more inference compute is required. We are starting with a version optimized for Pro users today, with up to 100 queries per month. Plus and Team users will get access next, followed by Enterprise. We are still working on bringing access to users in the United Kingdom, Switzerland, and the European Economic 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 still provides 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 to meet our release standards, we anticipate releasing deep research to Plus users in about a month.

Deep research is available today on ChatGPT web, and will be rolled out to mobile and desktop apps within the month. Currently, deep research can access the open web and any uploaded files. In the future, you’ll be able to connect to more specialized data sources—expanding its access to subscription-based or internal resources—to make its output even more robust and personalized.

Looking further ahead, we envision agentic experiences coming together in ChatGPT for asynchronous, real-world research and execution. The combination of deep research, which can perform asynchronous online investigation, and Operator, which can take real-world action, will enable ChatGPT to carry out increasingly sophisticated tasks for you.

_February 3, 2025 addendum: We conducted rigorous safety testing, preparedness evaluations, and governance reviews on the early version of o3 that powers deep research, identifying it asMedium_⁠(opens in a new window)_risk. We also ran additional safety testing to better understand incremental risks associated with deep research's ability to browse the web, and we have added new mitigations. We will continue to thoroughly test and closely monitor the current limited release. We will share our safety insights and safeguards for deep research in a system card when we widen access to Plus users._

Deep research unlocks significant new capabilities, but it’s still early and has limitations. It can sometimes hallucinate facts in responses or make incorrect inferences, though at a notably lower rate than existing ChatGPT models, according to internal evaluations. It may struggle with distinguishing authoritative information from rumors, and currently shows weakness in confidence calibration, often failing to convey uncertainty accurately. At launch, there may be minor formatting errors in reports and citations, and tasks may take longer to kick off. We expect all these issues to quickly improve with more usage and time.

Deep research in ChatGPT is currently very compute intensive. The longer it takes to research a query, the more inference compute is required. We are starting with a version optimized for Pro users today, with up to 100 queries per month. Plus and Team users will get access next, followed by Enterprise. We are still working on bringing access to users in the United Kingdom, Switzerland, and the European Economic 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 still provides 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 to meet our release standards, we anticipate releasing deep research to Plus users in about a month.

Deep research is available today on ChatGPT web, and will be rolled out to mobile and desktop apps within the month. Currently, deep research can access the open web and any uploaded files. In the future, you’ll be able to connect to more specialized data sources—expanding its access to subscription-based or internal resources—to make its output even more robust and personalized.

Looking further ahead, we envision agentic experiences coming together in ChatGPT for asynchronous, real-world research and execution. The combination of deep research, which can perform asynchronous online investigation, and Operator, which can take real-world action, will enable ChatGPT to carry out increasingly sophisticated tasks for you.

* * *

_February 3, 2025 addendum: We conducted rigorous safety testing, preparedness evaluations, and governance reviews on the early version of o3 that powers deep research, identifying it asMedium_⁠(opens in a new window)_risk. We also ran additional safety testing to better understand incremental risks associated with deep research's ability to browse the web, and we have added new mitigations. We will continue to thoroughly test and closely monitor the current limited release. We will share our safety insights and safeguards for deep research in a system card when we widen access to Plus users._