On the heels of its o3 mini release, OpenAI announced a new ChatGPT agent for handling in-depth research. Appropriately, it’s called “deep research.”
Using a yet-to-be-released version of OpenAI’s reasoning model o3, deep research “conducts multi-step research on the internet for complex tasks,” said the announcement. Deep research on ChatGPT essentially functions as your very own research assistant. It’s designed to help users “who do intensive knowledge work in areas like finance, science, policy, and engineering,” but can be used for any kind of thorough research needs, like shopping for a new car.
Tweet may have been deleted
ChatGPT’s deep research report on the best snowboards for freestyle riders, complete with sources.
Credit: OpenAI
If OpenAI’s deep research tool sounds familiar, that’s because Google Gemini released the same thing with the same name in early December. The two companies have continued to compete for AI dominance, taking that rivalry into the next phase of AI development, which is building models with more complex reasoning abilities and agents that can perform tasks on the user’s behalf. Then there’s the China-based DeepSeek, which exploded onto the scene last week with its cost-efficient R1 model, throttling tech stocks and spooking the AI industry.
OpenAI, Microsoft, Trump admin claim DeepSeek trained AI off stolen data
While R1 certainly made waves for being a fully capable LLM made for a fraction of the cost of ChatGPT, OpenAI seems to be flexing its muscles with the release of o3 mini, an agent called Operator for browsing the web and performing computer tasks, and now the deep research feature. “Deepseek’s r1 is an impressive model, particularly around what they’re able to deliver for the price,” OpenAI CEO Sam Altman posted on X last Monday. “We will obviously deliver much better models and also it’s legit invigorating to have a new competitor! we will pull up some releases.”
Mashable Light Speed
Tweet may have been deleted
Altman went on to say, “we are excited to continue to execute on our research roadmap and believe more compute is more important now than ever before to succeed at our mission,” in a follow up post referencing the company’s plans for AGI and the billions of dollars raised to ostensibly achieve this mission. So, OpenAI is at least publicly staying the course, but we can also expect some product releases to keep the masses happy and engaged.
Tweet may have been deleted
Compared to other models like GPT-4o, which is best for real-time conversations, deep research is best for specific and in-depth tasks, which can take anyway from five to 30 minutes to complete. As the announcement notes, you can step away from your computer while deep research does its thing, and you’ll get a notification of a final report within the chat when its done.
The responses include detailed outputs and summaries of its thinking, so you can see how deep research came to its conclusions. The report also includes citations and links to its sources.
Given the large compute power per query, OpenAI is first making deep research available to ChatGPT Pro users with 100 queries a month. Then Plus and Team users, then Enterprise users. Deep research is currently only available on the web, with mobile and desktop app access coming by the end of February. Sorry to users in the UK and EU, but OpenAI is “still working on bringing access” to you.
