Delegation Policy

Route tasks to the right model and agent for optimal quality per token.

Model Tiers (2026)

TierAnthropicOpenAIGoogleOpenCode GoCost
FastClaude Haiku 4.5GPT-5.4-miniGemini 3 FlashGLM-5, MiniMax M2.51x
BalancedClaude Sonnet 4.6GPT-5.4Gemini 3.1 ProKimi K2.5, MiniMax M2.73x
DeepClaude Opus 4.6GPT-5.3 Codex5x

Agent Routing

Agent CLIBest forDefault tier
claudeArchitecture, complex reasoning, multi-file refactorDeep (opus)
codexFast edits, scaffolding, mechanical changesBalanced (gpt-5.4-mini)
geminiLarge-context analysis, docs, code reviewBalanced (gemini pro)
aiderGit-native focused edits, test writingBalanced (configurable)
opencodeGeneral-purpose, prototypingBalanced (configurable)

When to escalate

  • Same bug investigated twice without root cause → Deep tier
  • Architecture decision with unclear trade-offs → Deep tier
  • Security implications span multiple systems → Deep tier
  • Need to search code or read logs → Fast tier
  • Need large codebase analysis → Gemini (large context)
  • Need many small focused edits → Codex or Aider

Target distribution

70% balanced, 20% fast, 10% deep = ~60% cost of all-deep with equivalent quality.