Better and cheaper together: Open models explore, frontier models patch

A multi-model pipeline can significantly reduce costs and improve performance on complex tasks by using cheaper models for initial exploration and a powerful frontier model for the final solution. This approach uses an executor-orchestrator architecture where 'junior' open models explore a codebase, a 'senior' model extracts relevant context from their attempts, and a 'principal' frontier model writes the final code patch. Using this method on the SWE-Bench Pro benchmark achieved a state-of-the-art 80.8% resolve rate while costing only $5.99 per task. The key to its efficiency is providing the expensive frontier model with highly refined, relevant context, allowing it to generate a correct patch in a single, focused attempt.