That's enough that you can generate 1000 messages a second for several millenia.
You'd want a probability more like 1-in-a-million (or 1-in-a-billion). That's 2^39 (or 2^34), which is 1000 messages a second for 17 years (or 7 months).
Again, probably still safe for many use cases. But you do have to think about it for a second. XChaCha20's 192-bit nonce capacity lets you totally ignore that factor, which is a nice property for generic crypto recommendations.
Generally speaking leaking that two plaintexts are the same is better than leaking two plaintexts XORed together. If you’re in a use case where that’s liable to happen (i.e., you don’t rekey “frequently”) I’d happily take the former leaked bit over the latter leaked messages.
There are also a handful of existing constructions you can just take off the shelf rather than rolling your own thing and getting it wrong.
With AES-GCM, if you reuse a nonce with two different plaintexts, you reveal the XOR of the two plaintexts!
This is less of a problem when each message also generates a random key, but still annoying to contend with.
XChaCha20-Poly1305 lets you generate 2^96 messages before having to worry about rekeying. https://libsodium.gitbook.io/doc/secret-key_cryptography/aea...