DeepSeek Flash as orchestrator for a 4-tier LLM routing stack

OpenCode as an alternative stack: routing agent work across 4 LLM tiers Running Codex and Claude Code for primary work. A separate stack for the rest: internal tooling, batch agents, experiments, the calls where paying frontier rates per request does not make sense. OpenCode is that stack after tuning. This setup runs alongside Codex or Claude Code, or standalone. It is boring on purpose. The core idea In my setup, most agent calls do not need a frontier model. They need a fast model for routing and classif