Candles themselves are simply a way of displaying the open, high, low and close for any specified time-period.
They do this with 100% accuracy, given an accurate price data-feed, for whatever time-frame they're set for.
In general, if you're looking at the extent to which their patterns can have any probabilistic predictive insight into future price-movements, "the longer the more reliable".
However you display the opens, highs, lows and closes (and I no longer use candles, myself), the display-patterns will always be more reliable, but give fewer "signals", on longer/slower timeframes.
If you identify something that has an "edge" in that the overall price-direction after a particular pattern will be anything other than 50/50, it's going to become more reliable as the periodicity of the candles gets longer, and less reliable as it gets shorter.
Note that that doesn't mean that the candle-patterns on longer/slower charts are necessarily more profitable, though, because profitability and reliability are two different parameters.
It may be that shorter/faster charts are still more profitable, because the inferior reliability of their "pattern signals" may be more than compensated for by the much greater trading-frequency they enable (and that's my experience of them).
It's more profitable to make an average of 15 pips profit with 70% reliability twelve times a day than it is to make an average of 25 pips profit with 80% reliability twice a day. (I'm throwing out nearly-random figures here, just to illustrate the point I'm making.)
In other words, when you're trading, don't confuse "win-rate" with "overall profitability". It's nice to have a system that's based on 75%/80% accuracy, but it's not much use if the occasional losses are really big, and it's also not much use if it's a system that only trades very rarely.