> For the complete documentation index, see [llms.txt](https://docs.dapta.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.dapta.ai/flow-studio/ai-nodes.md).

# AI Nodes

What the Flow Studio AI nodes do, the models each lab offers, how Dapta AI differs from your own credential, and what each run returns and costs.

An **AI node** is a Flow Studio step that sends a prompt to an AI model and hands its answer to the next steps: classify a message, summarize a call, extract fields from free text, read a PDF or an image, or draft a reply. There is one node per AI lab: **Google Gemini**, **OpenAI** and **Claude** (Anthropic). They all work the same way; only the models change.

This page is the overview. For the step-by-step setup of each node, open its own page:

{% content-ref url="/pages/fSdq5tTBO6lvs6J4crKW" %}
[Google Gemini Node](https://docs.dapta.ai/flow-studio/nodes/google-gemini-node.md)
{% endcontent-ref %}

{% content-ref url="/pages/M2QmTASRtALiwim2OAyR" %}
[OpenAI Node](https://docs.dapta.ai/flow-studio/nodes/openai-node.md)
{% endcontent-ref %}

{% content-ref url="/pages/llrYmUdtcz34IeaPv9rq" %}
[Claude Node](https://docs.dapta.ai/flow-studio/nodes/anthropic-node.md)
{% endcontent-ref %}

***

## Dapta AI or your own credential

The **Connection** section, the first one in the node, decides who runs the model and who bills it.

| Connection                 | Who bills the model                                                                                                                                | What you need                                         |
| -------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------- |
| **Dapta AI** (the default) | Dapta, in [Dapta credits](https://docs.dapta.ai/flow-studio/ai-nodes/credits.md), from November 10, 2026 (engine v2 flows: from November 13, 2026) | Nothing. It works out of the box.                     |
| **My own credential**      | Google, OpenAI or Anthropic, on your own account. It doesn't use Dapta credits.                                                                    | A credential with your key, added once per workspace. |

With **Dapta AI** you can also turn on **Authorize fallback**: if the model is busy or fails, the node retries with a **Backup model** you pick, from any lab. With your own credential the backup model is not available.

***

## The models

The **Model** menu shows these models for new nodes. The default is the model a new node starts with.

| Lab           | Models in the menu                                                                                                    | Default               |
| ------------- | --------------------------------------------------------------------------------------------------------------------- | --------------------- |
| Google Gemini | Gemini 3.5 Flash-Lite, Gemini 3.8 Flash, Gemini 4 Argon, Gemini 2.5 Flash-Lite                                        | Gemini 3.5 Flash-Lite |
| OpenAI        | GPT-6.1 Sol, GPT-6.1 Terra, GPT-5.6 Terra, GPT-6 Luna, GPT-5.6 Luna, and the open-source GPT-OSS 120B and GPT-OSS 20B | GPT-5.6 Terra         |
| Claude        | Claude Sonnet 5.5, Claude Opus 5.5, Claude Fable 5.1, Claude Haiku 5.5                                                | Claude Sonnet 5.5     |

Each model shows three meters (**reasoning**, **speed** and **price**) and whether it supports **Reasoning** and **Structured output**. Every model in the menu supports structured output. What each model costs in credits is in [AI node credits](https://docs.dapta.ai/flow-studio/ai-nodes/credits.md).

{% hint style="info" %}
A node saved on a model that is no longer in the menu keeps that model and keeps working. You see it selected in the node, and you can switch to a menu model at any time.
{% endhint %}

***

## What a run returns

**Output** decides the shape of the answer:

* **JSON (Recommended)**: the default. The answer arrives as JSON, ready for the next steps, with no Code node to parse it.
* **Text**: the answer as plain text.
* **Structured**: fields you define (name, type, required), filled in by the model.

**Response detail** decides how much the next steps receive:

* **Answer + details**: the default. The same fields for every model: `answer`, `json`, `usage` (tokens in, out, thinking and cached) and `costs` (the credits of the run). Type `{` in a later step to pick them, for example `{{your_node.answer}}`.
* **Raw response**: the provider's full response, unchanged. Its fields vary by lab.

***

## Defaults worth knowing

* **Attachments**: up to 5 files per run, 15 MB each and 20 MB in total: PDF, Word (.docx), PNG, JPG, WebP and GIF.
* **Max answer length**: empty by default, so the model can use its full length. The model only bills what it actually writes.
* **Creativity** (temperature): shown only for models that accept it.
* **Thinking** (Gemini), **Reasoning effort** (OpenAI) and **Extended thinking** (Claude): thinking makes complex tasks smarter, and the thinking is billed like answer text.
* **Authorize fallback**: off by default.

***

## Best practices

* **Start on the default.** Each lab's default (**Gemini 3.5 Flash-Lite**, **GPT-5.6 Terra**, **Claude Sonnet 5.5**) is the balanced pick between answer quality, speed and price. Build the flow on it and change the model only when a run shows why.
* **Go lighter for short, repetitive tasks.** Classifying a message, extracting a few fields or answering yes or no works well on the models with one price bar, such as **Gemini 2.5 Flash-Lite**, **GPT-6 Luna**, **GPT-5.6 Luna**, **Claude Haiku 5.5** or the GPT-OSS models. On a flow that runs thousands of times, the model is the biggest lever on cost.
* **Go bigger only when the default falls short.** Long documents, several attachments, multi-step reasoning or answers the default gets wrong are what the models with three reasoning bars are for: **Gemini 4 Argon**, **GPT-6.1 Sol**, **Claude Opus 5.5** and **Claude Fable 5.1**. They show fewer speed bars and more price bars. Turning on thinking on the default model is the step in between.
* **Keep JSON output when you switch models.** Every model in the menu supports structured output, so changing the model never changes the shape your next steps receive. Point later steps at `answer`, `json`, `usage` and `costs`, which are the same for every lab, rather than at **Raw response**, which changes when you change lab.
* **Use fallback for flows that must not stop.** With **Dapta AI**, turn on **Authorize fallback** and pick a **Backup model** from another lab, so one provider being busy does not stop the flow. **Last run** tells you when the backup answered.
* **Leave old nodes on their model until you edit them.** A node saved on a model that left the menu keeps working as it is. Switch it to a menu model when you are editing that flow anyway, then run it once and check **Last run**.

***

## Cost, caching and older nodes

{% content-ref url="/pages/A08OVFljHFrN7OWyWAdV" %}
[AI Node Credits](https://docs.dapta.ai/flow-studio/ai-nodes/credits.md)
{% endcontent-ref %}

{% content-ref url="/pages/o87AI8hBhlgm9xLDsyZc" %}
[Caching and Savings](https://docs.dapta.ai/flow-studio/ai-nodes/caching-and-savings.md)
{% endcontent-ref %}

{% content-ref url="/pages/VwwX2cOl3O9zEAyVixKR" %}
[Upgrade an Older AI Node](https://docs.dapta.ai/flow-studio/ai-nodes/upgrade-an-older-ai-node.md)
{% endcontent-ref %}

{% content-ref url="/pages/DoGCg7TMxXQOfuU05uU0" %}
[What's Next for AI Nodes](https://docs.dapta.ai/flow-studio/ai-nodes/whats-next.md)
{% endcontent-ref %}


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.dapta.ai/flow-studio/ai-nodes.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
