OpenAI to offer deep research agent for ChatGPT • The Register
OpenAI today launched deep research in ChatGPT, a new agent that takes a little longer to perform a deeper dive into the web to come up with a response to a query.
According to OpenAI, the new agent will “find, analyze, and synthesize hundreds of online sources to create a comprehensive report at the level of a research analyst.” It uses a version of the company’s upcoming o3 model to trawl the internet for information, pivoting as needed in reaction to what it encounters.
It can take anywhere from five to 30 minutes to complete its work. OpenAI claimed: “It accomplishes in tens of minutes what would take a human many hours.”
OpenAI published a plethora of statistics to back up its claims. On the Humanity’s Last Exam evaluation, a dataset of 3,000 questions across a hundred subjects designed to benchmark LLMs, OpenAI deep research managed an accuracy of 26.6 percent. By way of comparison, GPT-4o scored 3.3 percent, and Grok-2 managed 3.8 percent.
Users will be forgiven for experiencing a jolt of déjà vu. Google rolled out Deep Research to Gemini Advanced subscribers on December 11, 2024, and claimed the technology would save users “hours of time.”
Google’s Deep Research works by creating a multi-step research plan for a user to either revise or approve. Once given the go-ahead, the bot trawls the internet on the user’s behalf.
OpenAI’s deep research is more geared for asking ChatGPT a question, perhaps adding additional resources such as spreadsheets for context, and then letting it run. The result includes citations and a summary of how the agent came up with its response. However, the onus remains on the user to reference and verify the information returned by the software.
And verification continues to be necessary: OpenAI stated that inaccuracies and hallucinations occurred at a lower rate than existing ChatGPT models – according to the company’s internal evaluations. “It may struggle with distinguishing authoritative information from rumors, and currently shows weakness in confidence calibration, often failing to convey uncertainty accurately.”
The deep research agent is only available for Pro users, who pay the company $200 per month. Plus and Team users will be added next, followed by Enterprise. One hundred queries per month are permitted, although OpenAI said that paid customers would soon get “significantly higher rate limits” as the company releases faster versions powered by a small model.
The timing after the arrival of AI models from Chinese startup DeepSeek is interesting. DeepSeek has made claims about the models’ greater efficiencies and performance. As for OpenAI? “Deep research in ChatGPT is currently very compute intensive,” the US business said today.
OpenAI’s deep research agent is currently web-only, although there are plans to roll it out to mobile and desktop applications within the month. There is also the intent to allow customers to extend the agent’s reach by connecting it to more specialized data sources.
In the longer term, OpenAI envisages a combination of deep research and Operator, which can take real-world action, to “enable ChatGPT to carry out increasingly sophisticated tasks.” ®
