ChatGPT Deep Research

ChatGPT Deep Research is OpenAI's autonomous research agent, built into ChatGPT, that browses the web for 5 to 30 minutes to produce cited, comprehensive reports on complex topics. Rather than answering immediately it plans a research strategy, browses dozens of sources without step-by-step approval, and synthesizes the findings into a structured report with citations. It runs on OpenAI's current GPT-5 series models, with the faster instant model retained as the default. OpenAI changes the available model line-up often enough that any fixed version list dates quickly. Users track progress in real time and can interrupt to refine the plan or adjust which sources it may use. At launch on February 2, 2025 OpenAI reported that the model powering deep research scored 26.6% on Humanity's Last Exam and reached state of the art on the GAIA benchmark for autonomous web research. Those figures describe the original launch model rather than the models in use today. OpenAI names Bain & Company among the teams using it. Deep Research acts as an MCP client, letting users connect external data sources and restrict searches to trusted sites, which closes the public-web-only limitation of the original launch. By default it can use the public web, uploaded files and connected apps with read-only access. OpenAI now documents these under Apps rather than Connectors, and states that some apps can be used with deep research for multi-source analysis with citations back to the originals. Visual browser integration through agent mode extends research to pages that need visual understanding. OpenAI publishes no MCP server of its own for this product, re-verified August 2026. Pricing follows ChatGPT plan tiers and every figure is USD. OpenAI's own help center states that ChatGPT Plus is $20 per month billed monthly, and that annual billing is not supported for Plus at all. Free includes a limited number of deep research queries. Go at $8 per month adds messages and uploads but does not expand deep research beyond the free allowance. Plus at $20 per month is the entry tier that actually expands deep research. Pro from $100 per month provides the highest limits and priority access. Business is sold per user per month with admin controls, SAML SSO and no training on business data, and Enterprise is quote only, so both sit behind a sales conversation rather than self-serve checkout. What the $20 Plus tier does not buy is worth stating plainly, because it is where buyers are most often surprised. HIPAA eligibility is not available on Plus. OpenAI offers it only under a Business Associate Agreement, and only on ChatGPT for Healthcare, ChatGPT for Enterprise with Regulated Workspace, ChatGPT FedRAMP or ChatGPT for Clinicians. Deep Research is named as covered functionality on those products. Plus also does not include admin controls, SAML SSO, or the contractual commitment that business data is not used for training, all of which start at Business. Query caps apply on every consumer tier, and only Pro carries the highest limits. Who it is not for: Deep Research is web-first, so it is not the best fit for structured literature review of indexed academic databases. Elicit ($49/mo billed annually), Consensus ($12/mo billed annually) and ResearchRabbit ($10/mo) handle systematic reviews of PubMed, arXiv and Semantic Scholar better. Teams running high-volume research daily can also hit Plus query caps, which pushes them to Pro. Current state Q3 2026: reports can be downloaded as Markdown, Word or PDF. Independent review evidence for this row is a disclosed proxy: OpenAI publishes no separate review corpus for Deep Research, so the rating and review count shown here belong to the ChatGPT product profile rather than to Deep Research specifically. Security, verified on OpenAI's trust portal, which is scoped to ChatGPT services including Enterprise and Edu: SOC 2 Type II, ISO/IEC 27001:2022, ISO/IEC 42001:2023, GDPR and CCPA, plus HIPAA under a BAA on the products named above. ISO/IEC 27017, 27018 and 27701, CSA STAR, SOC 3, TX-RAMP and PCI DSS are also published there but sit outside this directory's approved certification list. ChatGPT Enterprise for FedRAMP has achieved FedRAMP Moderate accreditation through the FedRAMP 20x program, but OpenAI's published feature list for FedRAMP workspaces does not include deep research, so no FedRAMP claim is recorded on this row.