If the docs don't specify the cycle length, IMHO better not use it in situations where it matters.
65**5 is about 30 bits. if the PRNG-state is 32 bits, and the PRNG algorithm doesn't cycle through the full state, that may already be too little.
In many languages the default PRNG is implemented as a LCG, which are just 2-3 lines of code, and just not very good on their own. Computational science has found much, much better algorithms that are (almost) just as tiny to implement. Maybe 4-5 lines, see PCG-random.org.
I recently used a Java port of CMWC4096 from StackOverflow, for a numerical simulation in Java (where the default Math.random() has a stupid-bad cycle length, and the Java crypto-PRNG might have been too slow). It worked great. Not quite as tiny as PCG-random, but still short and it was ready-to-use Java-code that fit my needs.