Summary
OpenAI launched Deep Research in ChatGPT Pro on February 2, 2025, powered by an early version of o3. It scored 26.6% on Humanity’s Last Exam, according to OpenAI. That expert benchmark does not measure everyday report accuracy. Check its sources because Deep Research can make mistakes.
Introduction
OpenAI Deep Research is a ChatGPT feature for multi-step web research. It searches for information, analyzes sources, and creates a report with citations. OpenAI’s launch announcement described the feature and its limits; see OpenAI’s announcement.
At its February 2, 2025 launch, Deep Research used an early version of OpenAI’s o3 model. OpenAI optimized it for web browsing and data analysis. It belongs to the same model family as o3-mini, but the products serve different tasks.
What Deep Research did at launch
OpenAI said Deep Research could browse the web and read uploaded files. It could analyze text, images, and PDFs, then combine its findings in a report. Citations help readers inspect sources. They do not prove every statement is correct.
Launch benchmark results
OpenAI reported these Humanity’s Last Exam scores in its February 2025 launch announcement:
| Model | Score reported by OpenAI |
|---|---|
| Deep Research | 26.6% |
| DeepSeek R1 | 9.4% |
| OpenAI o3-mini (high) | 13.0% |
| OpenAI o3-mini (medium) | 10.5% |
| OpenAI o1 | 9.1% |
| Google Gemini Thinking | 6.2% |
| GPT-4o | 3.3% |
The test covers expert-level questions. The score is not a general accuracy rate and does not show how often an ordinary report will be correct. See the benchmark details.
How much did Deep Research cost at launch?
Deep Research first launched for ChatGPT Pro users. OpenAI launched ChatGPT Pro at $200 per month in December 2024. That was the plan price, not a per-report fee. Plan options have changed, so check OpenAI’s current pricing before subscribing.
Limits to keep in mind
OpenAI warned that Deep Research can hallucinate. It may confuse reliable sources with rumors. Open citations for important claims and check dates and numbers. Consult a qualified expert before making a high-stakes decision.
Why AI Entrepreneurs Should Care
Deep Research can help with tasks such as competitor scans or supplier comparisons. Treat its report as a first pass. Confirm key facts in linked sources before making a business decision. See our guide to AI tools for entrepreneurs for more options.
Final Thoughts
Deep Research can save research time, but its benchmark score does not replace a source check. Verify important claims before sharing or acting on a report.
For a separate model comparison, see our OpenAI vs DeepSeek overview. For an example research workflow, see the AI-powered research guide.
FAQ
When did OpenAI introduce Deep Research?
OpenAI announced Deep Research on February 2, 2025, as a ChatGPT feature powered by an early version of o3.
How much did Deep Research cost at launch?
OpenAI included Deep Research with ChatGPT Pro at launch. The plan cost $200 per month in December 2024, not per report. Check OpenAI’s pricing page for current plans and access.
How accurate are Deep Research reports?
OpenAI reported a 26.6% score on Humanity’s Last Exam for the model powering Deep Research. That score applies to one expert benchmark, not every report. OpenAI warned that the feature can hallucinate. Check important claims against its citations.
