Control how it answers
Choosing a model and what it costs
Every plan can use every model. What changes is what each answer takes out of your credits — one, three or six.
Before you start
- You know what a message credit is How to check
About 4 minutes. You’ll do three things:
- what each model costs an answer
- what that does to your month
- where to change it
Your bot answers with one model, and you pick which one. No plan hides a model from you — the list is the same on Free as it is on Advanced. The only thing your choice changes is how many credits each answer spends.
Every model you can pick
Nine models are on the list today, from four suppliers. The name tells you whose it is. The picker sorts them strongest first, and prints the price on every row.
| Model, as the picker names it | Credits per answer | What it suits |
|---|---|---|
| Claude Opus 4.7 | 6 | The strongest on the list, and the dearest answer you can buy. Worth trying when a 3-credit model still gets your hardest questions wrong. |
| GPT-5.5 | 6 | OpenAI's strongest. It works through a question before it replies, so answers start more slowly. |
| GPT-5.4 | 3 | A step up from the everyday OpenAI models, at three times the credits. |
| Claude Sonnet 4.6 | 3 | A step up from Haiku that keeps the Claude extras: product tables and Calendly times still work. |
| Perplexity Sonar Pro | 3 | A different supplier at the same price as the two above. Try it if you want to compare how an answer reads. |
| DeepSeek V4 Flash | 1 | An everyday model with room for very long pages of training material. |
| GPT-5.4 Mini | 1 | OpenAI's everyday model, at everyday price. |
| Claude Haiku 4.5 | 1 | What every new bot starts on. Cheapest tier, and it can use product tables and Calendly times. |
| GPT-5.4 Nano | 1 | The lightest model here. Best for short, simple answers. |
What the multiplier does to your month
Your plan buys credits, not answers. On a 1-credit model those are the same thing. On a 3-credit model they are not, and that is the whole trade.
Take the Local plan. It is £8 a month for 200 credits. On an everyday model that is about 200 answers. Move the same bot to a 3-credit model and those 200 credits buy about 66 answers. Nothing about the plan changed. Only what each answer takes out of it.
| Credits per answer | Answers on Local (200 credits) | Answers on Free (50 credits) |
|---|---|---|
| 1 | about 200 | about 50 |
| 3 | about 66 | about 16 |
| 6 | about 33 | about 8 |
There are five plans. Free is £0 for 50 credits, Local £8 for 200, Growth £20 for 500, Scale £75 for 4,000, and Advanced £250 for 14,000. On an everyday model, read each of those credit numbers as answers: about 50, 200, 500, 4,000 and 14,000 a month. Those are the pound prices. Your own currency, the annual option and your live balance sit on the pricing page and on Plan & billing.
Your own sum takes a minute. Count the questions you expect in a month, multiply by the credits your model costs, and pick the first plan above that number. A busier month on a cheap model beats a quiet one on an expensive model.
Where you change it
Change your bot's model
In the bot sidebar, under Bot setup, click Bot behaviour.
Click the Model & advanced tab.
Under Model, choose the model you want.
Every row prints its own price, like Claude Haiku 4.5 — uses 1 message credit per response.
Click Save.
Saved. Changes are live for new conversations. appears beside the button.
Two things your model also decides
| Feature | Which models it works on |
|---|---|
| Answering from a spreadsheet or product table | Claude models only. On any other model the tables upload fine and the bot ignores them. |
| Offering Calendly times inside the chat | Claude models only, for the same reason. |
| The Temperature slider | Every model except the four GPT ones. On a GPT model the slider moves and the model ignores it. |
| Everything you trained the bot on | Every model. Your website pages, documents and FAQs are not tied to a model. |
The tables page says the first one out loud. Upload a table to a bot on a GPT model and it shows Table queries require Claude. The warning points you at Settings → AI, which is what this page used to be called; Bot behaviour → Model & advanced is the same setting.
Temperature, and when it does nothing
Temperature sits under the model. The page explains it: Higher = more creative answers but less consistent. Customer support typically wants 0.0–0.3. The slider runs from 0 to 1 and names itself Reserved, Balanced or Creative as you drag it.
On the four GPT models the slider still moves and the model still ignores it. Those models answer at their own fixed setting, whatever the slider says. If you want the setting to do something, stay on a Claude, DeepSeek or Perplexity model.
If it didn’t work
The model I want is greyed out and I cannot choose it
- Check
- Read the end of the row. A greyed row ends with (coming soon), and Mistral Large is the only one today.
- Fix
- Pick another model from the list. There is no setting that turns a coming-soon model on, and no plan that includes it.
- Resolved when
- The Model box shows the model you picked and Save is available.
I uploaded a product table and the bot ignores it
- Check
- Open Bot behaviour → Model & advanced and read the Model box. If the name does not start with Claude, that is your answer. The tables page shows Table queries require Claude. for the same reason.
- Fix
- Choose Claude Haiku 4.5 for 1 credit an answer, or Claude Sonnet 4.6 for 3, then click Save. Ask the same question again.
- Resolved when
- The bot answers with a row out of your table.
I moved the temperature slider and nothing changed
- Check
- Read the Model box above it. The four GPT models ignore this setting.
- Fix
- Switch to a Claude, DeepSeek or Perplexity model and click Save if the setting matters to you. If it does not, leave the slider where it is — it is doing no harm.
- Resolved when
- Answers get looser or tighter as you move the slider and save.
My credits fall faster than the number of questions
- Check
- Open Bot behaviour → Model & advanced and read the credits on your model's row. A visitor who asks four questions gets four answers, and each one costs that much.
- Fix
- Choose a model whose row reads uses 1 message credit per response and click Save. Your training, your wording rules and your colours all stay as they are.
- Resolved when
- Usage moves by one for each answer instead of three or six.
Visitors saw "The assistant is temporarily unavailable. Please try again."
- Check
- That is the chat window's own line when the model cannot be reached. It appears with a Try again button, and your fallback message is not used.
- Fix
- Ask again in a minute — a supplier hiccup usually clears itself. If it keeps happening, choose a different model in Bot behaviour → Model & advanced, click Save, and email hello@adoomi.ai with the bot's name and roughly when it started.
- Resolved when
- A test question on your site gets a real answer.
What this can’t do
- Product tables and Calendly times work on Claude models only — on any other model the bot ignores them.
- The Temperature slider does nothing on the four GPT models. They answer at their own fixed setting.
- Mistral Large cannot be chosen. It is greyed out in the list and marked (coming soon).
- The model is set per bot, not per workspace — change it on each bot you run.
- Model cost on Analytics is our own estimated spend in US dollars, not your bill. Its own note says so: Estimated USD, last 30 days, all events. What you actually pay is your plan, and that is on Plan & billing.
- There is no per-question model switching. One model answers everything until you change it.