Cheaper, More Powerful Patch Generation Through Collaboration Between Open Models and Frontier Models
Key point
By using a tiered architecture where a cheap model extracts context and a frontier model writes the final patch, the approach captures both cost efficiency and performance at the same time.
Details
The existing executor-orchestrator structure had frontier models doing the planning while smaller models handled execution, but this research redesigned the entire pipeline to maximize cost savings.
The new architecture works like a team. The Junior model (MiniMax-M3) explores the repository to locate where changes are needed, the Senior model (GPT-5.2) organizes key context such as relevant functions and tests, and finally the Principal model (Opus 4.8 or Fable 5) writes the exact patch in a single call based on the prepared brief.
This strategy achieved an 80.8% resolve rate on the SWE-Bench Pro benchmark. Compared to third-party hybrid approaches, this result lowers the cost per task to around $5.99 without sacrificing quality.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.