Skip to content

Choosing your AI model: quick answers or in-depth answers

Simple or In-depth, GPT-6 Luna or Mistral Small: what really changes between the blabla.bot AI options, and how to choose the right one for each client site.

Costs and AI models5 min read

Faced with a list of AI models, the temptation is to take the newest or most powerful one, just in case. That is not always the right call. A heavier model answers more slowly, costs more, and adds nothing for a simple question like “Are you open on Sundays?”.

blabla.bot deliberately keeps its catalogue small: two models, each in two modes, so four options. This article explains what sets them apart and how to choose for each client site.

The four catalogue options

OptionModelModeNotes
GPT-6 Luna · SimpleGPT-6 Luna (OpenAI)SimpleDefault recommended option, the cheapest
GPT-6 Luna · ApprofondiGPT-6 Luna (OpenAI)In-depthRicher answers
Mistral Small · SimpleMistral Small (Mistral AI)SimpleSovereign option, announced EU hosting
Mistral Small · ApprofondiMistral Small (Mistral AI)In-depthSovereign option, richer answers

The option labels currently appear in French in the console (“Approfondi” means In-depth). GPT-6 Luna · Simple is preselected when you create an agent, and you can change it at any time.

What the mode changes: Simple or In-depth

In the current configuration, Simple and In-depth use the same model. What changes is the answer profile.

SettingSimpleIn-depth
Knowledge passages given to the model48
Maximum answer length500 tokens1,500 tokens

In practice:

  • In Simple mode, the agent draws on the four most relevant knowledge passages and gives a short answer. It suits factual questions: opening hours, prices, terms, lead times.
  • In In-depth mode, the agent looks at twice as many passages and can write a longer answer. It helps when the answer is spread over several pages or needs explaining: comparing two offers, detailing a procedure, explaining a warranty.

More passages and longer answers mean more tokens, so a higher cost and a slightly longer response time. These profiles are configuration settings, and their definition may evolve as we observe real usage.

What the model changes: GPT-6 Luna or Mistral Small

The two models differ in price and in provider.

  • GPT-6 Luna is the OpenAI model in the catalogue. With the prices checked on 13 November 2026, it is the cheapest option.
  • Mistral Small is Mistral AI’s model. It costs a little more. We present it as the sovereign option: Mistral AI states that it hosts data in the European Union by default. blabla.bot has not verified this hosting itself, hence the wording “announced EU hosting” in the console.

The console also shows, for each option, a relative speed and depth rated 1 to 4. These compare the four options with each other; they are not performance measurements.

What each option costs

On each option card, the console shows an estimated cost per conversation and per month at the agent’s current volume (conversations over the last 30 days). These estimates rest on catalogue assumptions: a 4-exchange conversation, with about 4,000 input and 250 output tokens per exchange in Simple mode, and 7,000 and 600 in In-depth mode.

With the prices checked on 13 November 2026 and blabla.bot’s ×2 multiplier, that gives:

OptionEstimated cost per conversation
GPT-6 Luna · Simple€0.0042
Mistral Small · Simple€0.0060
GPT-6 Luna · In-depth€0.0080
Mistral Small · In-depth€0.0113

An In-depth mode therefore costs roughly twice as much as the Simple mode of the same model. The gap is around half a cent per conversation, but it scales with volume.

How to choose, site by site

Here is a simple method.

  1. Start from the recommended option, GPT-6 Luna · Simple, unless the client has a specific constraint.
  2. If the client wants their data to stay in Europe as much as possible, choose a Mistral Small option. It is an explicit choice: blabla.bot never switches models automatically.
  3. Test with real questions in the “Test” simulator. Take the ten questions the client gets most often, plus two or three hard ones.
  4. Move to In-depth only if answers are too short or incomplete on the hard questions, and those questions are common on the site.
  5. Review the choice after a month, based on real conversations and the cost shown in the console.

Some pointers by type of site:

  • Local shop, restaurant, tradesperson: short, factual questions; Simple mode is usually enough.
  • Online store: Simple for delivery and returns questions; In-depth if questions are about choosing between many products.
  • Consultancy, software publisher, training provider: answers are often longer, so In-depth may be worth it.

Changing option along the way

A change of option applies from the agent’s next answer: there is nothing to reinstall on the site. You can start cautiously and adjust.

Remember to tell your client about a model change, especially if they asked for the sovereign option: it is a commitment you made to them, not just a technical setting.

Worth knowing: the answer model does not change how knowledge is indexed. Passage retrieval uses the same service for every agent, whatever option is chosen.

In short

Choose the model based on what the client expects (price, or announced hosting in Europe), and the mode based on how complex the questions are. Start with the simplest, test, then adjust.

To compare the monthly cost of the four options at your sites’ volume, use the calculator on the Pricing page.

Sources

Tags

  • AI models
  • costs
  • configuration