Assuming that only a brute-force attack can be preformed and you are asking about current timescales, the answer is that for symmetric ciphers like AES with select key spaces, it is almost certainly the 128-256 bit scale (for now.)
The main thing is principles like Landauer's Principle which would make the pure energy made to brute-force these keys so enormous. If this were to somehow be bypassed in the next hundred years or further, we could just make an enourmous key like 512-bits or 1024-bits before those become weak. DES with 56-bits is still capped with a brute-force time of 24 hours which was set by David Hulton and Moxie Marlinspike. We are very, very far.
With cryptosystems like RSA and Elliptic Curves, this becomes a bit more tricky to define. The best we have for RSA is the general number sieve, which runs in sub-exponential time. A weaker RSA-892 bits is the current record for the highest RSA broken. NIST has put various estimates on when RSA-2048 and higher will be broken, with the current estimate being 2030. Brute-force would be stupid expensive to even run.
The general case for RSA and other cryptosystems that rely on prime numbers is that we will have to increase it until increasing it is no longer necessary.
For Elliptic curves, brute-force is the same situation with AES and most other ciphers that aren't reliant on primes and modular arithmetic.