Models and credits
Which model writes the answers, what an answer costs, and what happens when the credits run out.
Choosing a model
Every agent has one model, chosen under Settings or in the Playground. New agents start on gpt-oss-120b, an open model that is fast, inexpensive and good at answering from given text. For most agents there is no reason to change it.
The list comes from our model provider and is refreshed hourly, so it changes over time. Models are grouped by maker, and beside each one is what an answer costs.
A larger model does not know your business any better. It reads the same passages. What it can add is better handling of long, subtle or multi-part questions, and a more natural tone. Compare before you switch: the Playground puts two models side by side on the same question.
What an answer costs
You pay in credits, and the price is per answer, not per word. An answer costs the same whether it is one line or five paragraphs. Most models cost 1 credit an answer. Larger models cost more, up to several credits.
| Plan | Credits a month |
|---|---|
| Free | 100 |
| Starter | 1,500 |
| Studio | 4,000 |
| Agency | 12,000 |
| Enterprise | unlimited |
Credits are shared by all agents in the workspace and start again on the first of each month (UTC). Unused credits do not carry over.
Good to know:
- Test questions count. A Playground question costs the same as a visitor's. Comparing two models costs both.
- The second opinion is free. The grounding check is not charged to you.
- A greeting counts. "Hello" gets an answer, and an answer costs a credit.
- An answer is charged once it is written, also when the visitor closes the chat halfway through.
- Answers are capped in length, so one question can never run up a large bill.
- A file costs extra. A document or a picture a visitor sends with a question costs 2 credits on top of the answer on nearly every model, and more on the dearest. Only a model with a picture mark in the list looks at pictures. A page of a scan read by text recognition costs 1 credit more. Files from visitors.
Models on the Free plan
Free runs every model that costs 1 credit an answer. Models that cost more are shown with a padlock. Every paid plan can use all of them.
When credits run out
When the month's credits are used up, the agent stops answering until the next month, or until you make room. A visitor who asks something at that point is told that the agent cannot answer right now and to try again later. It does not mention credits, and your visitor is still left with nothing, so it is worth preventing:
- Watch usage under Settings, where the month's credits are shown.
- Move up a plan if you reach the limit regularly.
- Buy credits (paid plans) under Settings, in packs of 1,000 for €30. They are added as soon as the payment is confirmed, and they do not expire.
- Switch on auto-recharge (paid plans). When your remaining credits fall below a level you choose, a pack of 1,000 credits (€30) is added and put on your next invoice. Purchased credits do not expire.
Plans and credits has the details.
Temperature
Temperature sets how freely the model words its answers, from 0 to 1. New agents start at 0.70.
- Lower (0.2 to 0.4) stays closer to your sources and answers the same question the same way. Right for support, policies and anything with numbers in it.
- Higher varies more and embellishes sooner. Fine for a brand with a playful voice, risky for facts.
A few models only accept their own fixed temperature. For those the setting is ignored and the answer is written anyway.