Definition
What is reasoning effort in a coding agent?
Reasoning effort is a setting that controls how much a model thinks before it answers. Higher effort produces better results on hard problems and uses more tokens and more time. On mechanical work it usually buys nothing, which is why it is worth being able to change it per turn rather than per session.
Last updated · Markdown version
A rough rule
| Work | Effort |
|---|---|
| Rename, reformat, mechanical edit | Low |
| Write a feature in a familiar codebase | Medium |
| Debug something nobody understands yet | High |
| Design a migration with real consequences | High |
Why per turn matters
A thread rarely stays at one difficulty. Being able to raise the effort for the one hard step and drop it again afterwards is cheaper than running the whole session at the maximum, and a cost view makes the difference concrete rather than theoretical.
Frequently asked questions
Does higher effort always help?
No. On mechanical work it usually just costs more tokens and more time.
Can I change it mid thread?
In Helicon, yes. The model and reasoning effort picker applies from the next turn.
How do I see what it cost?
The usage page breaks cost down by model, which is where the difference shows up.
Keep reading
- Keyboard and paletteA command palette, slash commands and a model and reasoning effort picker. Every project, session and action in Helicon is reachable without the mouse.
- Cost at API ratesSee what every Muse Code thread would have cost at published per token rates, broken down by day, by model and by thread. A meter reading on your subscription, not a bill.
- See what a thread costOpen the usage page to see what every thread would have cost at published per token rates, by day, by model and by thread, plus your plan meter.
Same Muse Code. Same subscription. Better interface.
Free and MIT licensed. Signed Windows installer, universal macOS DMG, and a web build against a daemon you run.