Is Computational Indistinguishability quantifiable?

The answer is yes. Formally, when a scheme is proven secure (relatively to a specific security game), a theorem states that no adversary $mathcal{A}$ couldn't win the security game except with a probability $f(lambda, m_1, m_2, dots)$.

With $f$ a function in $[0; 1]$, $lambda$ a parameter called the security parameter (in general when $lambda$ increases $f$ decreases as the efficiency (in time and space) of your protocol) and $m_1, dots m_n$ which determines (or just upper bound) the power capability of the adversary: For example $m_1$ could be the number of messages seen by the adversary. And $m_2$ could be the number of generic group operation (if we are in the generic group model).

You have first to determine $m_i$ (id est: If your adversary is your nephew, the NSA, your nephew in 2050, the NSA in 3850, etc). Then you choose lambda which verifies:

$f(lambda, m_1, m_2, dots) <frac{1}{2^{128}}$.

Particular case: Some result contain terms like $texttt{ADV}^P(lambda)$, with $P$ a well known computational problem (RSA, DH, LWE, etc). Then you have to check in the literature the best known attack to evaluate this term (or more simply which lambda is recommended).