This study evaluates a blind Binaural Speech Intelligibility Model (bBSIM) that predicts speech intelligibility (SI) and listening effort (LE) for spatially separated target speech and noise interferers. The model applies a blind binaural front-end, which assumes that the interfering noise has different modulation characteristics than speech and which predicts the effects of better-ear-listening and binaural unmasking. This binaural front-end is combined with two different blind back-ends that estimate SI and/or LE based on the output signal of the binaural front-end without any further auxiliary information. These blind back-ends are the non-intrusive Short Time Objective Intelligibility measure (NI-STOI) and an automatic triphone recognizer with an a posteriori estimate of the certainty of the recognized triphones. Even a short-term version is available which can be applied to arbitrary unknown speech and which updates SI and LE estimates in close to real time. This offers new fields of application, e.g., in hearing instruments. Therefore, the blind real-time BSIM has been implemented on a research hearing aid, where it continuously estimates SI and LE. These estimates can be used, for instance, for informing the selection of hearing aid processing strategies.