Plan with Claude Opus, Build with Kimi K2.6? LIVE Mixed-Provider Benchmark

Last Monday I added Opus into the Dark Factory's workflows for deep reasoning while keeping Kimi K2.6 as the workhorse for everything else. It works - the factory shipped its first mixed-provider PR cleanly. But I never measured WHICH mix was actually optimal. Should Opus plan and Kimi implement? Or Kimi plan and Opus implement? Today I'm answering that with data. Using Archon, I built a benchmark matrix that runs the same coding workflow four different ways - varying which provider handles planning and which handles implementation - against three real GitHub issues. After all 12 PRs land, an Opus-based evaluator scores each one on seven dimensions (root cause, scope discipline, subtle correctness, code quality, test discipline, plan-to-impl fidelity, solves-the-issue). 🔗 Archon: https://archon.diy 🔗 Dark Factory: https://github.com/coleam00/dark-factory-experiment 🔗 Dynamous community: https://dynamous.ai Livestreams Mon/Thu/Sat at 9 AM CT. New YouTube video every Wednesday at 7 PM CT.