OpenAI’s latest model is a personal research assistant

OpenAI on Sunday unveiled Deep Research, a new AI agent that’s capable of conducting complex, multi-step online research into a variety of topics (a DeepSeek, if you will).

How it works: In response to a single query, Deep Research can search the web and analyze massive amounts of text, images, and PDFs to compile a detailed, fully cited report “at the level of a research analyst.” According to OpenAI, it “accomplishes in tens of minutes what would take a human many hours.”

• Deep Research is intended for professionals in finance, science, engineering, and similar disciplines – but it can also be used by consumers to analyze major purchases like cars and furniture, per OpenAI.
• It marks the second AI agent OpenAI has released this year, alongside Operator. Both models are available to subscribers of ChatGPT’s new $200/month Pro tier, with future plans to make them more widely available.

But…Deep Research is still subject to the same Achilles heel that affects all other AI models – sometimes, it just makes sh*t up. OpenAI says its latest tool has an infrequent tendency to hallucinate facts and present users with incorrect information, albeit at a “notably” lower rate than ChatGPT (based on internal testing).

📸 Big picture: Deep Research’s launch comes as OpenAI has committed to speeding up its AI product releases in response to the recent advances made by DeepSeek. OpenAI is also reconsidering its overall development approach and could switch to making its models open source like DeepSeek’s, meaning anyone could use them for free and alter their underlying code, CEO Sam Altman revealed on Friday – a somewhat ironic potential occurrence, given OpenAI is in the midst of transitioning away from a non-profit structure and to a for-profit one.