All posts
Guide· 1 min read

Picking the right model for each phase of a task

Claude, Codex, or Kimi — and at what effort? A practical guide to per-phase routing in .trau.ini for cost and quality.

By The trau herd

Picking the right model for each phase of a task

Not every phase of a task needs your most expensive model. Planning might want a strong reasoner; boilerplate edits do not. trau lets you assign a model and an effort level per phase.

Configuring .trau.ini

# per-phase routing: <PROVIDER>_<PHASE>_MODEL / _EFFORT
CLAUDE_BUILD_MODEL=fable
CLAUDE_BUILD_EFFORT=max

CLAUDE_VERIFY_MODEL=claude-sonnet-5
CLAUDE_VERIFY_EFFORT=medium

CODEX_COMMIT_MODEL=gpt-5.4-mini
CODEX_COMMIT_EFFORT=low

Match effort to the work and you control both cost and quality without babysitting the run. This is a placeholder; the full guide with cost tables is on the way.