OpenAI's ChatGPT Unveils a Major Upgrade

The company has just introduced a new agent for its flagship artificial intelligence product called “deep research.” The tool can scan the internet for data and generate reports with the same quality as a professional research analyst.

In a post on February 2, OpenAI said “deep research” is designed for professionals working in fields such as finance, science, policy, and engineering – industries that require thorough, accurate, and trustworthy research.

In addition, the company also emphasized that the tool is useful for researching information before buying expensive items such as cars or home appliances. Reports generated by “deep research” can take between 5 and 30 minutes and always come with full documentation, clear citations, and a summary of the analysis.

The launch of the tool comes shortly after OpenAI unveiled Operator on January 23, a ChatGPT agent that can use the internet to perform tasks like ordering groceries or planning a trip.

The event also coincides with a sharp upheaval in the AI market and US tech stocks following the emergence of DeepSeek, a new AI model from China that is said to be as efficient as ChatGPT but developed at a significantly lower cost.

An Example Of Chatgpt's Deep Dive Mode Showing Its Progress And Citations — An Example Of Chatgpt’s Deep Dive Mode Showing Its Progress And Citations

Microsoft and OpenAI are now investigating whether data from ChatGPT’s API was improperly collected by a group affiliated with DeepSeek.

OpenAI said its deep learning agent achieved a new record in the AI assessment called Humanity’s Last Exam, a test of 3,000 expert-level questions covering more than 100 topics. The tool achieved 26.6% accuracy, far surpassing the 9.4% of DeepSeek-R-1 and the 3.3% of OpenAI’s own GPT-4o model.

The agent is powered by the OpenAI o3 model, but has been “optimized for web browsing and data analysis.” OpenAI o3 is the company’s latest thinking model, designed to self-verify information to reduce errors and avoid creating inaccurate information.

However, OpenAI also warns that deep research can sometimes “hallucinate” information or make false inferences, and can have difficulty distinguishing reliable sources from rumors.

Last month, Google also announced plans to launch a similar feature, also called Deep Research, for its Gemini AI model in early 2025. Meanwhile, OpenAI said its research agent is now available on its $200/month Pro plan, with a limit of 100 queries per month.