OpenAI is adjusting its strategy for ChatGPT’s Deep Research tool, introducing a secondary, “lightweight” version aimed at providing faster, more cost-effective reports. The new option utilizes a version of the recently released o4-mini reasoning model. It began rolling out immediately not only to paying subscribers across Plus, Team, and Pro tiers but also, significantly, became available to users on ChatGPT’s free tier starting the same day, offering a more accessible path to structured AI research.
Understanding the Lightweight Approach
The standard Deep Research feature, primarily powered by the larger o3 model, is known for compiling detailed reports by scouring the web and synthesizing information from multiple sources, a process that OpenAI previously indicated could take between 5 and 30 minutes depending on complexity.
In contrast, the new lightweight version aims for speed and efficiency. OpenAI positions the o4-mini variant as “nearly as intelligent as the deep research people already know and love, while being significantly cheaper to serve,” resulting in responses that are “typically shorter while maintaining the depth and quality you’ve come to expect.” This cost-effectiveness is the key factor enabling OpenAI to now offer limited access to free users for the first time.
New Tiers and Limits for Research
The launch reconfigures access and usage allowances across the board. According to updated OpenAI help documents, free users receive 5 lightweight deep research tasks per month.
ChatGPT Plus and Team subscribers now get 10 standard Deep Research tasks plus an additional 15 lightweight tasks monthly. Users on the $200/month Pro tier, which originally had exclusive access when Deep Research debuted, see their allowance adjusted to 125 standard and 125 lightweight tasks per month.
Enterprise and Edu users are set to receive access matching the Team tier (10 standard + 15 lightweight) starting the week of April 28th. The system automatically defaults users to the lightweight version once they hit their monthly limit for standard Deep Research tasks, with limits resetting every 30 days based on first use.
Model Capabilities and Caveats
The underlying o4-mini model was part of OpenAI’s mid-April rollout of its new o-series, characterized by what the company terms “early agentic behavior.” These models can autonomously select and use tools like web browsing or code execution.
However, subsequent analysis raised questions about their reliability. OpenAI’s own data, reflected in the o3 and o4-mini system card, alongside external research, indicateS higher rates of hallucination or fabricating information compared to earlier models. OpenAI acknowledged in its system card that “more research is needed” on this front.
A Response to Competitive Pressures
OpenAI’s introduction of a more accessible Deep Research tier arrives amidst a rapidly crowding field of AI research assistants. Google notably made its competing Deep Research feature in Gemini completely free for all users on March 13.
Other players like Perplexity AI (with its $20/month real-time web focus), xAI’s Grok 3 (“Deep Search” for subscribers), and Anthropic’s Claude (adding agentic research for paid users) have all established similar capabilities.
Even Microsoft has added Deep Research now to its Copilot AI assistant. This competitive activity likely influenced OpenAI’s decision to broaden access beyond its initial high-cost Pro-only offering from February 2025 and the subsequent expansion to Plus users later that month. While offering wider access, the tool’s effectiveness currently remains limited to the open web and user-uploaded files, as it cannot yet access private data sources, though OpenAI notes this capability is planned *”in the near future.”*
Last Updated on May 2, 2025 1:36 pm CEST
