Smart vs Fast, and how much effort to ask for
Smart and Fast are two modes in the prompt bar, and effort is a separate dial. Learn which to use when, and what each costs.
Two controls in the prompt bar shape how the agent works on a request:
- Smart / Fast chooses which AI models do the work.
- Effort (Quick, Balanced, Thorough) chooses how much thinking and checking they do.
They are independent. You can ask for Fast with Thorough effort, or Smart with Quick effort.
Fast
Fast uses lighter, quicker models. It is the right default for most day-to-day work:
- simple edits (a fillet, a resize, a new hole),
- changing parameters,
- first drafts of ordinary shapes,
- questions and measurements.
It costs fewer credits and answers sooner.
Smart
Smart uses the most capable models available. It is worth the extra credits when the request is hard:
- multi-step builds from a prompt ("a box with a hinged lid, a snap latch and vent slots"),
- turning photos, sketches or drawings into models,
- tricky geometry where parts interact (fits, mechanisms, threads),
- requests that Fast got wrong twice.
Effort
| Effort | Best for |
|---|---|
| Quick | Obvious edits where you can spot a mistake at a glance |
| Balanced (default) | Most requests |
| Thorough | Anything that must be right: fits between parts, mechanisms, load-bearing shapes, requests with many constraints |
Higher effort means the agent plans more carefully and double-checks more. It takes longer, and it can use more credits.
What things cost
Credit costs depend on the action and the mode. The defaults at the time of writing:
| Action | Fast | Smart |
|---|---|---|
| Routing (working out what you meant) | Free | Free |
| Simple edit (fillet, resize) | 1 | 2 |
| Multi-step build from a prompt | about 6 | about 15 |
| Concept image (each) | 5 | 8 |
| Photo or drawing to model | 10 | 20 |
| Auto drawing sheet | 5 | 8 |
| Verification check (render and compare to your request) | Free | 1 |
Routing is always free, so a question like "can I print this?" is cheap to ask. Larger requests can cost more than the typical figure. Your balance and the amount charged are shown with each result, and Credits, explained has the full table.
A sensible habit
- Draft in Fast. Get the shape roughly right with cheap iterations.
- Refine the hard part in Smart. When the request involves fit, motion, or interpreting an image, switch.
- Use Thorough for the final fit-critical pass. Not for every message.
- Use Quick for tiny tweaks you will check visually anyway.
Try it
Run the same request in both modes and compare the ghost previews and the credit charge:
Add a lid to this box that sits flush, with a 1.5 mm lip that locates it and 0.25 mm clearance per side.
Use Reject after the first run, switch modes, and send it again. Compare how the two proposals handle the lip and the clearance, then keep whichever you prefer.
Did this help?
Keep reading
Account and billing
Credits, explained
What AI credits pay for, the typical cost of each action in Fast and Smart mode, what is free, and how to keep an eye on your balance.
Beginner4 min readStart here
Talking to the agent: text, voice, images and pointing
Four ways to tell the agent what you want, when to use each, and how they combine. The prompt bar is always at the bottom center.
Beginner4 min readTroubleshooting
The agent misunderstood me
The preview is not what you meant. Find out whether the cause was selection, missing numbers, a mixed-up request or the wrong mode, and recover without losing work.
Beginner5 min read