AAC is likely slightly better, though at bitrates found on Youtube, there is not a large difference, so you should choose the format based on other requirements like compatibility.
While it is true that all video and audio on Youtube are transcoded, Opus does not gain an advantage from this as indicated by the current answer. Repeated transcodes using the same codec does not reduce quality as much as you would expect from a normal transcode. Since the encoder tends to make the same or similar decisions about what data to throw away, a second pass will actually tend to keep the same data that was preserved during the first pass, resulting in fewer differences between repeated transcodes.
Here is one test demonstrating this phenomenon: After 100-passes of AAC transcodes, the result is surprisingly high quality. Additionally, IIRC on certain codecs, if you repeatedly transcode with the same settings, quality will amazingly not drop below a certain threshold, reaching an asymptote that depends on bitrate.
Since most video software encodes the audio track as AAC, the AAC track Youtube provides is likely an AAC -> AAC transcode, which will be better than an AAC -> Opus transcode.
Additionally, Opus is not always the best option for music. Although Opus is very good at a variety of audio, it does not necessarily beat all other lossy codecs in every case. Opus was originally designed as a low bit rate, low latency, speech codec for telephony applications. And while many incremental improvements were made so that it could deal with a much larger variety of audio, it still shows weaknesses in a few areas as a result of that initial design decision.
For example, Opus has an Achilles' heel where tonal music can be glaringly bad. In the 64kbps listening test, it failed miserably on a harpsichord sample (#2) against AAC. In the 96kbps listening test, it failed again on a guitar sample (take_your_finger_from_my_hair) and was even beat by MP3! Opus was originally designed with a short MDCT window for low latency, so it occasionally shows odd problems in places where AAC doesn't. The short window creates low frequency resolution that must be compensated by increasing the bitrate in order to get good quality on tonal sounds, implemented in the encoder with tonality estimation.