Multicomponent thermosets can simultaneously achieve high strength, modulus, and toughness inaccessible by single-component materials, but efficiently designing those with exceptional mechanical properties relies on serendipity because knowledge-guided methodologies often fail in high-dimensional spaces and multi-objective problems. We proposed a multi-objective Bayesian-optimization-based active learning method enhanced by a dual-driver acquisition function to accelerate the formulation design of multicomponent epoxy resins. After several iterative experiments, we identified a multicomponent formulation of ultra-strong and high-tough samples that was much better than currently available epoxy resins. The molecular dynamic simulations and synchrotron radiation wide-angle X-ray scattering revealed that the exceptional mechanical properties of multicomponent thermosetting polymers originate from the larger meshes of crosslinking networks and the homogeneity of cured structures. The closed-loop, dual-driver Bayesian optimization strategy can be generalized for the fast discovery of high-performance materials that involve time-intensive experiments and multi-dimensional parameter spaces.