Hmm I would have expected PDFs to work fine. The way gateways determine the mime is “magic” I believe, which can behave unexpectedly. The related code appears to be here. It looks like you’d need to access them like /ipfs/CID/1.pdf so it knows they’re actually PDFs (and passes the correct mime to the browser). It’s possible specifying ?filename=1.pdf for example to the tail of the URL would solve it too, but that’s not obvious to me from the code.