Last-mile pharmaceutical distribution must reconcile logistics cost, delivery responsiveness, and cold-chain safety under uncertain demand. This study develops a bi-objective two-stage stochastic vehicle-routing model that integrates direct home delivery with consolidated delivery through smart parcel lockers. Vehicle reservation and locker activation are decided before demand is observed, whereas vehicle dispatch, routing, channel assignment, service timing, and multi-product flows adapt to each scenario. The model distinguishes cold-chain from ordinary products, enforces direct delivery for temperature-sensitive items, and captures a heterogeneous fleet, simultaneous delivery and pickup, locker capacity, coverage limits, and time-window-based customer satisfaction. The augmented ε-constraint (AEC) method generates exact efficient solutions for tractable instances, while a scenario-aware NSGA-II solves larger cases. Across five common benchmarks, NSGA-II is approximately 4.1 times faster; across 15 paired instances it produces 34.5% more non-dominated solutions on average, while AEC retains a significantly lower mean ideal distance. A real pharmaceutical-distribution case in Tehran with 250 customers shows that an all-direct policy increases total cost by 11% and decreases customer satisfaction by 13.17% relative to the mixed strategy. The results also support a locker coverage radius no greater than 3 km and demonstrate substantial resilience to locker- and distance-cost changes in operationally realistic urban logistics settings.