Reported by @soft nebula
Bug Report: Inference failure between 5.2-Thinking and Deep Research
`Steps to Reproduce`
It's probably not reproducible as-is, but shows a real issue.
What I did :
- create a project with title "Game : Rune Factory Frontier on Wii", and description "All discussions in this project are exclusively related to Rune Factory Frontier"
- inside the project, a new chat session with modes "5.2 Thinking" + "Deep Research", asking for a manual for "this" game (+ a few more structural specs)
`Expected Result`
The Deep Research report provides an actual manual for the said game.
`Actual Result`
The Deep Research report provides a GENERIC manual for an unspecified game (which is incredible by itself but totally useless at the moment).
Here is the chat session (French) : https://chatgpt.com/share/698ccc19-5a74-8001-af38-79b0da862387
`Environment`
ChatGPT Windows app from the Microsoft Store (up to date as of Feb 11, 2026)