There’s no two ways about it, there’s a newfound sense of urgency at OpenAI. Two days after releasing o3-mini to the world, the company made a surprise announcement on Sunday evening, revealing Deep Research. The new feature allows ChatGPT to find, analyze and synthesize hundreds of websites and online sources to create reports “at the level of a research analyst.”
On top of the usual text questions, users can upload files, including PDFs and spreadsheets, when prompting ChatGPT in this way. The chatbot will then take “anywhere from 5 to 30 minutes” to compile an answer, a side panel documenting the agent’s progress and citations as it works. “It accomplishes in tens of minutes what would take a human many hours,” OpenAI says of the new feature.
“Our ultimate aspiration is a model that can uncover and discover new knowledge for itself,” said Mark Chen, chief research officer at OpenAI, during the company’s reveal livestream. “It’s core to our [artificial general intelligence] roadmap.”
As far as limitations go, OpenAI says ChatGPT can sometimes hallucinate facts or make incorrect inferences when conducting Deep Research, though “at a notably lower” rate than other current models. Additionally, the agent may sometimes struggle to differentiate between authoritative information and rumors. Users may also notice some formatting errors. “We expect all these issues to quickly improve with more usage and time,” the company notes.
If all of this sounds familiar, it’s because Google’s Advanced suite includes its own Deep Research feature, which not only shares the same name but broadly offers the same set of capabilities as well. One significant difference between the two is that Google offers access to Gemini Advanced through its $20 per month One AI Premium plan. By contrast, you’ll need a $200 per month ChatGPT Pro plan to start using OpenAI’s version of Deep Research today.
“Deep research in ChatGPT is currently very compute intensive,” the company reasons, adding it will limit Pro users to 100 queries per month. “The longer it takes to research a query, the more inference compute is required.”
OpenAI says it’s working on a version of Deep Research powered by a smaller, more cost-effective model. In turn, that will allow the company to offer “significantly higher rate limits.” In the meantime, OpenAI hopes to get the tool in the hands of Plus users “in about a month,” following a round of safety testing. As with most of the company’s other recent releases, European users will need to wait before they can try out the tool for themselves, with Deep Research not yet available to people in the UK, Switzerland and the broader European economic zone.