- After the Operator AI agent, OpenAI has added a new Deep Research AI agent for ChatGPT Pro users.
- It can go to the web and perform research for you independently. The agent takes 5 to 30 minutes to generate a comprehensive report.
- The Deep Research AI agent is built on the new o3 reasoning model. With web access, it scored 26.6% on Humanity's Last Exam.
Another day, another AI agent. After the Operator AI agent, OpenAI has now released a new AI agent called ‘Deep Research’ on ChatGPT that goes to the web to do “research” for you independently. OpenAI says the Deep Research AI agent on ChatGPT can browse the web, find accurate sources, analyze them, and create a comprehensive report for you.
It can take 5 to 30 minutes to perform deep research for you. ChatGPT’s Deep Research agent is very similar to Gemini’s Deep Research feature. The most noteworthy change is that Deep Research is based on the new OpenAI o3 model for web browsing and Python analysis. Thanks to the advanced reasoning capability of the o3 model, Deep Research can intelligently analyze PDFs, images, and text on the web.
As a result, when o3 is given web access (+ Python tools) to do deep research, it scores 26.6% on Humanity’s Last Exam — a new rigorous benchmark consisting of challenging questions from domain experts around the world. o3-mini-high without web access only scores 13% and the new DeepSeek R1 model gets 9.4%.
Currently, Deep Research is only available to users (100 queries per month) who are subscribed to the ChatGPT Pro plan, which costs $200 per month. OpenAI says Deep Research is coming soon to ChatGPT Plus (10 queries per month), Team, Enterprise, and free users too.
OpenAI further says Deep Research is “built for people who do intensive knowledge work in areas like finance, science, policy & engineering and need thorough & reliable research.” With the release of Deep Research, OpenAI CEO Sam Altman suggested that this AI agent can perform “a single-digit percentage of all economically valuable tasks in the world“.
This statement aligns with OpenAI’s definition of AGI in its charter — “highly autonomous systems that outperform humans at most economically valuable work.” It seems OpenAI is hinting that we are coming close to achieving AGI. Now we need to check how accurate and reliable the new Deep Research agent is. So stay tuned for our analysis.
