Introducing deep research

OpenAI just shipped something it's calling deep research, and the pitch is simple: stop asking ChatGPT questions and start handing it assignments. Instead of a quick back-and-forth, you give it a research task, and it goes off, reads a pile of sources, reasons through what it finds, and comes back with a structured writeup — the kind of thing that used to eat an analyst's afternoon.

The mechanism is what's interesting here. This isn't a chatbot doing one search and summarizing the top three links. It's built to chain steps together, pull from multiple sites, cross-check details, and adjust its search path based on what it's already learned, the same way a person digging into a topic keeps refining their next Google query based on the last one. OpenAI is framing it as an agent rather than an assistant, which is a distinction the company has been pushing hard since it started talking about 2025 as the year of agents.

Rollout is staged, which by now is standard OpenAI playbook. Pro subscribers get it first, Plus and Team users are told to expect it shortly after. That tiered release lets OpenAI stress-test the feature against paying power users before it hits a much bigger and more demanding audience, and it also nudges casual users toward the pricier tier if they want the new toy immediately.

What's notable is the target use case. This isn't for trivia or quick fact lookups — it's aimed at the kind of multi-step, source-heavy research that shows up in consulting decks, academic literature reviews, or due diligence work. If it actually holds up on accuracy and citation quality at scale, it starts to compete less with search engines and more with the entry-level grunt work junior analysts and research assistants currently get paid to do.