Most likely the key here is web search. Later I will try the post’s example with gpt-5 with search. I would be surprised if it didn’t say the same thing.

From a product point of view, it seems clear that just as they have work to get the model to dynamically decide to use reasoning when it would help, they have to do the same with web search.

gpt-5 with medium reasoning, no search: "Short answer: use the system zstd C library that ships with iOS; Apple’s Compression framework still doesn’t expose zstd. Link libzstd and call its streaming API from Swift." (and then like 50 lines of code plus some explanation)

gpt-5 with medium reasoning + search: "Short answer: you can’t. Apple’s built‑in Compression framework on iOS exposes LZFSE, LZ4, ZLIB/DEFLATE, and LZMA—not Zstandard—so there is no system API to do zstd without bundling an external library. (https://www.hackingwithswift.com/example-code/system/how-to-...)

If you can switch to a built‑in algorithm, here’s a minimal streaming compressor using Compression (LZFSE shown):" (etc etc)