LaTeX rendering in DeepResearch output

The LaTeX rendering in DeepResearch answers is almost always broken. I like the feature a lot, but it makes it hard to read the output for Math questions.

See e.g. below for an example where even simple formulas like $x$ get bad rendering:

Totally agree.


I agree. It’s still quite broken and makes reading the responses very difficult. And strangely, if I then just switch to the normal GPTPro and ask it to transcribe the prior response but fixing any latex escaping characters and whatnot, it totally fails and just writes a very small fraction of the originally generated deep research response. Almost like it can’t access the original information or something.

Did anyone find a solution?