Can you tell me how many hands it takes to make the main HUD stats reliable?

It depends on the stat. Something like VPIP will converge pretty quickly. Lots of other stuff which only becomes relevant every 50 hands will take longer to get the sample size needed.

Vague questions yield vague answers.

Preflop stats will converge quite close within 100-1000 hands. (VPIP/PFR/3B/F3B)

You're right
I'm newbie, play NL2 and I do a little bit of confusion. In my HUD I have
VPIP/PFR/3B/F3B
CBET/FCBET
ASTEAL/FSTEAL
I think, maybe I'm wrong, that 300 hands may be enough for VPIP and PFR, but for the others? Is it better to refer to opportunities than hands? How many opportunities to 3bet i.e do I have to have for the %3bet to be significant.
Ty

opportunities would probably be better to look at, and have samples 100+

Also should be noted that something like 3b% might converge in a relatively short amount of hands (100-1000 range), but if you're looking for a more specific 3B% such as BB vs SB 3b% then obviously because that's for a specific circumstance you're going to need a lot of hands.

Thanks

The answer is "the more the better."

The reason is simple: people are inconsistent. A particular opponent may be tilted today, drunk tomorrow, and a sober crusher the rest of the week. If you form solid opinions based on your first 300 hands, you may find your reads to be unreliable.

Your confidence in your stats should grow as they do. Treat few stats with a high level of suspicion, then become more confident as your database grows. There is, unfortunately, no magic number.

A coping mechanism is to substitute population tendencies for a particular opponent's stats. Find player information gleaned from the large pool of users in your vicinity and play against those models. Adjust later as needed.

Keep in mind that good players are adjusting to you as you learn about them. That guy who just 3bet-shoved K5s *might* be a maniac, or he might have seen you fold to 3bets 10 times in a row and is exploiting you.