OpenAI has announced the broader rollout of its Deep Research artificial intelligence (AI) agent, extending access to all users of the ChatGPT platform. In a shift from its previous availability exclusively to paid subscribers, the San Francisco-based company is now introducing a “lightweight” iteration of the tool that operates on the o4-mini AI model, albeit with specific usage limits.
OpenAI’s Deep Research AI Agent Available to All Users
The official OpenAI account shared details of this expansion in a post on X, formerly known as Twitter. Free users will gain access to the AI agent for the first time, while paid subscribers will benefit from enhanced query limits. The introduction of the lightweight model aims to make the tool more cost-effective, according to the company.
ChatGPT free users will now have the ability to perform five lightweight Deep Research queries each month, as noted on the company’s website. For ChatGPT Pro users, query limits have increased from 100 to 250 per month. Once users reach their allocated limit, the system will automatically revert to the lightweight version.
Additionally, users subscribed to Plus and Team plans can now conduct 25 Deep Research queries monthly. Enterprise and educational subscribers will gain access to the feature next week and will experience rate limits similar to those of the Team plan.
Deep Research in ChatGPT
According to OpenAI, the lightweight format of Deep Research offers intelligence that is “nearly as intelligent” as its standard counterpart, while also being more economical. Although users can expect shorter responses compared to the regular version, the company asserts that the quality will remain uncompromised.
Members of the Gadgets 360 staff who tested the lightweight version noted that an icon appears beneath the text field, adjacent to the reasoning icon. When using the AI agent, users must provide a primary query. ChatGPT will then ask follow-up questions to refine the focus of the research and better grasp the user’s needs. The research process is estimated to take between five and 30 minutes.
The entire sequence of reasoning is displayed in a sidebar on the right side of the interface, with sources cited alongside relevant sentences as well as listed at the bottom of the response.
