- OpenAI has launched Deep Research, a new ChatGPT tool that automates multi-step internet research.
- The agentic AI tool generates “expert-level” reports and is available to Pro users.
- The launch comes as competition heats up for OpenAI from China’s DeepSeek and domestic AI rivals.
OpenAI has launched Deep Research, a tool for automating complex multi-step internet research, as the company continues rolling out new products in the face of competition from Chinese startup DeepSeek.
In a Sunday blog post, OpenAI said Deep Research can operate independently to gather information from the internet to meet a user’s research needs, completing tasks in “tens of minutes that would take a human many hours.”
The AI agent, which is available for Pro users paying $200 a month, is aimed at professionals in finance, science, policy, and engineering.
Unlike other ChatGPT tools, Deep Research operates independently for five to 30 minutes, adjusting its approach in real time as it gathers information.
Deep Research is powered by a fine-tuned version of OpenAI’s o3 reasoning model. The tool autonomously browses the internet, pulls data from multiple sources, and produces fully cited reports, OpenAI said. Mark Chen, OpenAI’s chief research officer, called it “something that an analyst or an expert in a field might produce.”
According to OpenAI, Deep Research has already achieved a new high score of 26.6% on “Humanity’s Last Exam,” an AI benchmark of expert-level questions, beating GPT-4’s 3.3% score. However, OpenAI cautions the model can still “hallucinate” incorrect facts and can struggle to distinguish between rumor and fact.
In an X post on Sunday, OpenAI CEO Sam Altman encouraged users to “try it on your hardest work task that can be solved just by using the internet and see what happens.”
On Friday, OpenAI launched o3-mini, a cost-efficient reasoning AI model. The two OpenAI launches come just days after Altman said his company would deliver “much better models” and move more quickly in response to the release of DeepSeek’s R1 model.
DeepSeek shook the tech industry and Wall Street last month with a model that appears to match the capabilities of those from OpenAI but seemingly at a lower cost.
Last month, OpenAI rolled out its Operator AI agent, while in December it launched the text-to-video model Sora to the public.
