.svg){kind=logo}
ChatGPT Deep Research
AI system that autonomously browses the web
From Wikipedia, the free encyclopedia
Deep Research is an AI agent integrated into ChatGPT,[1][2] which generates cited reports on a user-specified topic by autonomously browsing the web for 5 to 30 minutes.[3]
| Deep Research | |
|---|---|
| |
| Developer | OpenAI |
| Initial release | February 3, 2025 |
| Type | AI agent |
| Website | openai |
Agent
Deep Research can interpret and analyze text, images, and PDFs.[4] Since February 2026, it uses a model based on OpenAI's GPT-5.2,[5][6] having originally been released with a specialized version of OpenAI's o3.[4]
o3-based Deep Research scored 26.6% on the "Humanity's Last Exam" benchmark, outperforming rivals like DeepSeek's model R1 (9.4%) and GPT-4o (3.3%).[7]
According to OpenAI, Deep Research occasionally makes factual hallucinations (errors) or incorrect inferences.[4] It may also reference rumors, and may not accurately convey uncertainty.[8]
In April 2025, OpenAI announced a "lightweight" version of Deep Research that would be available to free users, based on o4-mini instead of o3.[9]
As of June 2025, ChatGPT Pro subscribers ($200/month) receive 250 queries per month (half of which are "lightweight"), Plus, Team and Enterprise users receive 25 queries per month (15 of which being "lightweight"), and free users receive 5 "lightweight" queries per month.[10]
In February 2026, OpenAI announced updates to Deep Research with a new GPT-5.2-based model, better steering, limiting scope to select sites, connecting additional data using MCP servers, along with better UI for the final reports.[5][6]
Usage limit
OpenAI introduced "Deep Research", an autonomous research mode inside ChatGPT, in February 2025.[11][12] According to the company's support documentation, ChatGPT Plus, Team and Enterprise subscribers are allotted 10 "full-model" Deep Research queries every 30 days, followed by 15 additional queries on a cheaper "lightweight" model; Pro subscribers receive 125 + 125 queries, while free users get 5 lightweight queries.[10][failed verification] These figures have been corroborated by independent technology outlets covering the feature's rollout.[13][9][14] The quota is not displayed proactively in the ChatGPT interface; users are informed of the limit only after exhausting their allowance, at which point a banner states when new queries "will be available again."[15][16] Tech journalists and users have criticized this reactive disclosure, arguing that the uncertainty incentivises heavy users to upgrade to the US$200-per-month Pro tier that offers larger allowances.[17][18]