From the pricing theory, processes need to be martingales when divided by the numeraire asset.
A classical example is a stock option: Consider a money market $B$ being the numeraire asset. When we price a stock option with a payoff $h(S(T))$, then the money-market discounted stock price process $S/B$ has to be a martingale under the martingale measure associated with $B$.
But now consider a bond option where the bond's price is driven by and risk-free rate $r$ subject to a Vasicek process (under risk-neutral measure). The payoff of the bond option is $h(r(T))$. If we consider the dynamics of $r$ under the risk-neutral measure, $dr(t)=k(\theta - r(t))dt + \sigma dW^Q(t)$, then $r/B$ will clearly not be a martingale under $Q$.
My question is: How come that the discounted risk-free rate $r/B$ doesn't need to be a martingale under $Q$ if the stock had to?
I do understand that the discounted bond price in Vasicek model is a martingale under $Q$ but why the same doesn't apply to the risk-free rate in the bond option case?