Tune Effort Per Prompt

Effort controls how hard the underlying model thinks before answering. Higher effort costs more and runs slower but produces better output on complex tasks. Orbit lets you override the tier default per prompt or per benchmark candidate.

Set effort in a prompt’s routing tags

Add an effort: tag on the third line of your prompt header, after complexity::

[Orbit] #NNNN — Refactor the auth middleware
complexity: high
effort: high

Valid values are low, medium, high, xhigh, and max. When present, Orbit passes the effort through to the runner:

• Claude Code: receives --effort <value> in the spawn arguments
• Codex: receives the matching reasoning-effort config value

If you omit effort:, the runner uses the default effort level configured for that complexity tier in Settings → Model Tiers & Agents.

Change the tier default in Settings

1. Open Settings → Model Tiers & Agents.
2. Find the tier you want to adjust (Low, Medium, High, or Ludicrous).
3. Set the Effort field for that tier. All prompts routed to that tier will use this effort unless individually overridden.

Set effort per candidate in Benchmark Mode

In an agentic benchmark job, each runner row has its own Tier / Effort fields:

1. Open Benchmark Mode and create or edit a job.
2. In the builder, add or select a runner row.
3. Set the Effort field on that row independently of the other rows.

This lets you run the same prompt at medium effort on one row and high effort on another, then compare pass rate and cost in the results.

Tips

• Start with the tier default unless you have a specific reason to override.
• Use effort: low for documentation or formatting tasks where speed matters more than depth.
• Use effort: high or xhigh for architectural refactors, security analysis, or anything where a wrong answer is expensive to fix.

Related

• Complexity Tiers & Routing
• Run a Benchmark Job
• Queue & Approve a Prompt