You have to be careful: the implied constant also depends on $\chi$ (not just on $\epsilon$), in particular it depends on the number field $K$. (I edited your post to reflect this.)
There is a better (subconvex) bound available, namely there is an absolute constant $\lambda<1/2$ such that $$ L(\sigma+it,\chi)\ll_{\epsilon,K} C(\chi|\cdot|^{it})^{\lambda(1-\sigma+\epsilon)},\qquad 1/2\leq\sigma\leq 1, $$ where $C(\chi|\cdot|^{it})$ is the analytic conductor of the Hecke character $\chi|\cdot|^{it}$. Here $|\cdot|$ stands for the norm of ideals or ideles, depending on how you think of Hecke characters (Grössencharacters or idele class characters). This follows from the Phragmén-Lindelöf convexity principle combined with Theorem 5.1 in this paper of Michel and Venkatesh. In particular, $$\alpha=\lambda[K:\mathbb{Q}](1-\Re(s))$$ is admissible in your bound with the same absolute constant $\lambda<1/2$.
Added. The state-of-the art seems to be in Han Wu's thesis, see Theorem 0.3.4 there. According to this result, $\lambda=(5+2\theta)/12$ is available, where $\theta=7/64$ is the current record towards the Ramanujan-Selberg conjecture on $GL(2)$ over $K$. Perhaps the Burgess-like exponent $\lambda=(3+2\theta)/8$ is also within reach, especially in the light of the other main results of the thesis (for twists of cusp forms on $GL(2)$ over $K$), which have appeared in GAFA recently. See here.