# AI models

The platform connects models from various developers. Some are stronger and more expensive, others faster and cheaper; some recognize images and can reason, others cannot. There is one catalogue for your whole account: the same model can be chosen in the [AI chat](https://support.watbot.org/ai/chat) and in the [AI Agent](https://support.watbot.org/chatbots/blocks/ai-agent) block.

## Where to find it

The "AI models" section is in the project menu. It has three pages, with the menu on the right:

- "Models" — the catalogue;
- "Statistics" — token usage charts;
- "Logs" — a list of calls to models with their cost.

"Statistics" and "Logs" are described on the [AI Credits](https://support.watbot.org/ai/balance) page. A team member sees them only if their role allows it.

## Catalogue

The "Models" list contains all available models. Above the list are the "Provider" filter and the "Search" field by name.

A model row shows:

- the developer's icon and the name;
- what the model can do — the "Text generation", "Built-in reasoning", "Embedding" badges;
- the price for input tokens and the price for output tokens — per million tokens, in your account's currency;
- "Context" — how many tokens the model holds in one request.

Clicking a row opens the model page.

## Model page

The "About the model" page has a description and characteristics:

- **"Modalities"** — what the model accepts and what it returns: text, images, documents, audio, video.
- **"Input/output token price"** — per million tokens.
- **"Cached input"** and **"Cache write"** — separate prices for the repeated part of a request, if the model has them. What this means is explained on the [AI Credits](https://support.watbot.org/ai/balance) page.
- **"Context"** and **"Release date"**.
- Capability tags: "Text generation", "Built-in reasoning", "Tool calling", "Coding", "Structured output", "Multilingual".

The "Statistics" page shows token usage and the number of requests for this particular model in your project.

## How models differ

| Characteristic | What it affects |
|---|---|
| Token price | How much each response costs. Input is your question together with the conversation history, output is the model's response |
| Context | How much text the model sees at once: a long conversation, a large document or a big bot scenario needs a large context |
| Image recognition | Whether you can attach pictures. Models without it will not accept a picture |
| Built-in reasoning | Whether the model can "think" before answering. The level is set in the chat with the "Reasoning" item |

A price table for all models is on the [Prices of additional services](https://support.watbot.org/billing/services) page.

## How to choose

- **Everyday questions, texts, translations, replying to customers by instructions** — an inexpensive fast model. The difference in quality on such tasks is hard to notice, while the difference in price is easy to see.
- **Complex analysis, long documents, multi-step tasks** — a strong model with a large context; turn on "Reasoning" if needed.
- **Pictures and screenshots** — a model that accepts images.
- **Building a chatbot** — not all models are suitable. Unsuitable ones are marked in the model list while building is on in the chat, and on an unsuitable model the chat itself will suggest a replacement. See the [Building a bot in the AI chat](https://support.watbot.org/chatbots/ai-builder) page for details.
- **An AI agent in a bot** — it responds many times a day, so the model price matters most here. Start with an inexpensive model and check the responses with "Test the bot".

> [!TIP]
> You can ask different models the same question: change the model and click "Regenerate" under the response. The cost of each attempt is visible in "Logs".

## Favorites and the default model

Add the models you use most often to your favorites — with the star in the "All models" list in the chat. Favorite models are at the top of the model selection menu.

A new conversation in the AI chat opens on the first model from your favorites. Each user has their own favorites; on the first visit, a few models have already been added.

A conversation opened earlier continues on the model that gave the last response in it.

## Russian models

Some of the developers are Russian. Two things are specific to them:

- Building a chatbot is not available on Russian models — while building is on in the chat, the model list marks them "Chatbot building is not available on Russian models".
- When a conversation is on a Russian model, the chat's service requests — for example, the conversation name — are also performed by a Russian model.

## If a model has disappeared from the list

Models are updated over time, and outdated ones are removed from the catalogue. Conversations that were using such a model remain: when you open one, the chat will say "The model "…" is no longer available — the chat continues on …" and continue on the default model.