Deep Search represents a new AI tool developed by OpenAI which focuses on conducting intensive web searches. The tool performs sequential internet research according to Reuters reports to execute complex tasks.
The deep search engine operates through OpenAI O3 model v1 that was specifically developed to process web content and analyze data. The technology allows users to submit requests to ChatGPT which subsequently conducts text analysis and summary with PDF and image examination before generating research-compliant written reports.
OpenAI announced through its blog that this tool caters to researchers and scientists from fields such as engineering, politics and finance who demand trustworthy and extensive research findings. This tool provides assistance to people who need to conduct thorough investigations ahead of buying major household items as well as cars and appliances.
The "Deep Search" starts its availability with Pro users who receive 100 search queries per month period before making it available to additional member types. This tool will start to function for the Plus and Team users and eventually become available to Enterprise users during the following month.
The Deep Search tool functions through two steps: first choose the Deep Search option from the message box followed by entering your query. The system provides users with both a file attachment option and spreadsheet feature. A deep search using this tool will require between five minutes and thirty minutes to provide a response about your question which will generate a notification when the search finishes. After the search operation completes the sidebar shows an overview with both the steps and the referenced sources.
The tool operates through the web interface for now but developers plan to release mobile and desktop versions in the upcoming month.
How accurate is the search in "Deep Search"?
OpenAI declared that this new tool finishes operations within ten minutes which normally need hundreds of minutes for a regular person to execute. The tool remains in an early developmental phase so it should not surprise us that it faces several difficulties. The system demonstrates problems recognizing genuine reliable data from mere false information. The current state of the tool displays two prominent limitations including uncertain confidence levels together with unreliable information detection problems.
All outputs generated by "Deep Search" in ChatGPT are fully documented with clear sources while showcasing the core idea to enhance reference and verification capabilities. The company failed to establish whether their current measures would effectively counteract AI errors since AI web search tools often deliver faulty results. "ChatGPT Search" generated search results that proved inferior to Google results based on a TechCrunch assessment.
OpenAI enhances deep search accuracy through the deployment of "O3" model version which underwent reinforcement learning training for browser and Python language-based real-world operations.
OpenAI performed "Deep Search" evaluations with "Humanity’s Last Exam" because it contains more than 3,000 expert-level questions from diverse academic subjects. Similar to other deep search operations the O3 model reached accuracy of 26.6%. The seemingly low "Humanity’s Last Exam" accuracy rate exists because its creators made it more difficult to spur AI model advancement.
The company's statements indicate that this accuracy level stands significantly higher than the rates found in "Gemini Thinking" at 6.2% and "Grok-2" at 3.8% and "GPT-4o" at 3.3%.
Any user concerned about generative AI's influence can prefer deep outputs with proper sources since these provide more authenticity than undocumented anonymous summaries. We must observe how users approach these outputs since they might choose to utilize them as professional content without genuine examination.
