Reasoning effort
Published Updated
What is Reasoning effort?
A setting that controls how much a model thinks before answering; more effort generally takes longer and uses more output tokens.
Reasoning effort controls how much a model thinks before it answers. More effort can improve careful multi-step work, but it is slower and consumes more tokens; thinking tokens are billed as output even when they are hidden. Claude Code offers `/effort` with low, medium, high, xhigh, and max. Codex selects effort in `/model`. Codex's top setting, Ultra, is reached through `/model` → a model → More reasoning… and is labeled “highest usage”; it can hand parts of a job to helper agents running at the same time, and it is not offered on GPT-6 Luna. Claude Code's `ultracode` is a separate switch, not an effort level: `/effort ultracode` for a session, or the word `ultracode` in one prompt, runs the job as a multi-agent workflow at the effort level already set and uses more of your limits.
How it relates to Orbit
Each Orbit Model Tier sets its own effort: Default, Low, Medium, High, Extra High, or Max.
When you would use it
Adjust reasoning effort when you need to trade response speed and token use against the care a task receives.
Common operations
- Choosing low effort for quick, well-bounded answers.
- Choosing more effort when a coding or reasoning task needs additional deliberation.