Deep Research

Deep Research entered the ToTheMoon conversation in early 2025 and quickly became a working tool for the hosts rather than a headline. They discuss which tasks justify waiting for a long report, where it loses to an ordinary search, and why the user has to pick a mode by hand at all. Through Deep Research the channel asks a broader question: how to tell a strong model from a usable product.

Episode 044 described the release as an analytical tool that behaves like a person, searching on its own and asking clarifying questions, and cited a test where it scored 26.6% accuracy against 3.3% for 4o. The limits showed within a month: the mode is useful for a complex market study, but on an everyday request such as finding a dog trainer the output was no better than Google or Yelp (episode 048). In episode 050 one host admitted he was happy to wait ten or twenty minutes, even hours, for a report. The real value was seen in long tasks such as vetting a person across hundreds of sources, though separating fact from guesswork still falls to the user (episode 052).

By summer 2025 the discussion had moved to the interface. Choosing between GPT-4o, o3 and Deep Research is done by hand, while a mature product ought to decide the mode itself (episode 061); the ChatGPT agent does exactly that, and in a side-by-side test its output looked much the same (episode 068). In 2026 the verdict got tougher: the report can be strong, but the clarifying questions, delays and shifting quality grate, and users now compare the whole working process rather than a single answer (episode 096).