In this work, a physics-based crystal plasticity model is developed to predict failure in Grade 91 steel. A microstructure-sensitive dislocation kinetics law defines local plastic slip, an Arrhenius creep law is used to model vacancy-mediated plasticity, and strain hardening evolves with local dislocation density. As voids nucleate, a reaction–diffusion framework is adopted to dynamically track the local void size distributions, which grow by coupled viscoplastic and diffusive processes. Upon accurately reproducing the temperature and stress dependencies in primary, secondary, and tertiary creep seen experimentally for Grade 91 steel, the model is exercised to generate a material response database across a wide range of operating conditions. A new reduced-order lifetime predictor is developed from numerical predictions, and a Bayesian framework is used to quantify prediction uncertainties. When compared to current empirically-derived lifetime relations, the proposed lifetime assessment tool predicts rupture times up to several orders more conservative.