We were all expecting OpenAI to announce two new next-gen ChatGPT models on Friday to meet the self-imposed late January deadline. OpenAI delivered, unveiling its new o3-mini and o3-mini-high reasoning models. The o3 versions are already available to ChatGPT users, and the names are indeed getting out of hand here.
I mention product names before anything else because OpenAI surprised the world on Sunday with a product we were not expecting. Well, we didn’t expect another AI agent to drop so soon after the Operator and o3-mini releases, though ChatGPT’s new Deep Research AI agent is something any fan of AI products has been waiting for. I know I was.
It also sounds a lot like DeepSeek, the Chinese AI that rocked the tech world last week. And yes, Google has its own Deep Research feature built into Gemini Advanced, which works just like OpenAI’s.
I often use ChatGPT for various research, telling the AI to find me complex information. I use it for work and personal projects that would otherwise involve multiple separate internet searches and a lot of time spent on different tabs to collect the data I need.
The basic version of ChatGPT (GPT-4o with ChatGPT Search abilities) can get the job done. But it’s not always perfect. It might need massaging and repeat attempts. It’s not always as thorough as I want it to be. Also, I still look at sources to check whether the AI has hallucinated certain things. That’s where a better ChatGPT reasoning model could come in handy.
That’s exactly what Deep Research is, and it’s available now in ChatGPT. It’s a reasoning model based on the unreleased o3 model, which takes longer to scour the web for accurate information. Deep Research then delivers everything into a larger report, which could be much more useful than anything GPT-4o can offer.
As the name might suggest, Deep Research will take a lot longer to offer an answer. You might have to wait several minutes for your report rather than mere seconds, which is the case with current ChatGPT models. It can take anywhere from 5 to 30 minutes to offer responses, at which point ChatGPT will deliver a notification.
You can move to other chats with the AI in the meantime and perform different tasks while you wait for a Deep Research report.
Deep Research is currently available with a $200/month ChatGPT Pro subscription. You’ll get 100 queries per month at first, as performing these longer tasks is “very computer intensive.” OpenAI explains, “The longer it takes to research a query, the more inference compute is required.”
The first Deep Research version will be optimized for Pro users. ChatGPT Plus and Team users will get it next, with Enterprise to follow. Unfortunately, ChatGPT users in the EU, like myself and those in the UK and Switzerland, will have to wait longer for this new OpenAI model.
How does Deep Research work? OpenAI says the new AI model can look at information in real-time in ways other models can’t. The AI will look at the data and adapt as it goes along:
Deep Research is OpenAI’s next agent that can do work for you independently—you give it a prompt, and ChatGPT will find, analyze, and synthesize hundreds of online sources to create a comprehensive report at the level of a research analyst. Powered by a version of the upcoming OpenAI o3 model that’s optimized for web browsing and data analysis, it leverages reasoning to search, interpret, and analyze massive amounts of text, images, and PDFs on the internet, pivoting as needed in reaction to information it encounters.
As for the training model, OpenAIz used end-to-end reinforcement learning, similar to what the DeepSeek researchers did with their R1 model:
Deep Research was trained using end-to-end reinforcement learning on hard browsing and reasoning tasks across a range of domains. Through that training, it learned to plan and execute a multi-step trajectory to find the data it needs, backtracking and reacting to real-time information where necessary. The model is also able to browse over user uploaded files, plot and iterate on graphs using the python tool, embed both generated graphs and images from websites in its responses, and cite specific sentences or passages from its sources. As a result of this training, it reaches new highs on a number of public evaluations focused on real-world problems.
The resulting reports will contain plenty of links to document every claim. Deep Research results should make it easier to check the facts, though OpenAI says ChatGPT will hallucinate less in this mode. Still, the hallucination element can be present, albeit “at a notably lower rate than existing ChatGPT models, according to internal evaluations.”
In the future, the reports will pull visual data from the web, including images and graphs.
For now, Deep Research tops existing benchmarks, including Humanity’s Last Exam, GAIA, and internal OpenAI tests.
All of this only makes me want to try ChatGPT Deep Research and integrate it into my workflow. Unfortunately, I’ll have to wait a while.
For actual Deep Research examples, check out OpenAI’s blog post at this link. Even better is the video below, which shows two ChatGPT Deep Research demos performed in real-time during Sunday’s surprise livestream.
