> ## Documentation Index
> Fetch the complete documentation index at: https://docs.quarkle.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Models & providers

> The AI models behind Quarkle, how Auto chooses one, and what Pro model selection unlocks.

Quarkle isn't tied to a single AI model. It routes each task to a model that's good at it, across several providers. **Pro** users can override that choice in [Chat](/ai/quarkle-chat).

<Frame caption="Pro users can pick a model, or leave it on Auto.">
  <img src="https://mintcdn.com/quarkle/5WUUaYl-dgOEAFoM/images/app/chat-models.jpg?fit=max&auto=format&n=5WUUaYl-dgOEAFoM&q=85&s=dceaed3bf68df114fd85d89e4c80b3aa" alt="The Quarkle model picker" width="1074" height="720" data-path="images/app/chat-models.jpg" />
</Frame>

## Overview

Quarkle picks a model per task by default, from a menu of frontier models across several providers. Leave it on **Auto** and you never have to think about it. Pro users can pin a specific model instead.

## Auto: the default

Leave the model on **Auto** and Quarkle picks one for you, per task. It weighs a few things: what you're doing (a quick chat reply, a deep brainstorm, or a JSON-structured review), how big the context is, and which models are healthy and fast right now. If a provider is having a bad moment, Quarkle fails over to another one automatically. For most writing, Auto is the right choice.

## Choosing a model (Pro)

Pro users can pick a specific model from the **#** menu in Chat. The line-up evolves as new models ship. Today it includes frontier models such as:

|                                                  |                                    |
| ------------------------------------------------ | ---------------------------------- |
| **Claude Sonnet** (Anthropic)                    | **GPT-5.4** and **GPT-5** (OpenAI) |
| **Gemini 3.1 Pro** / **Gemini 3 Flash** (Google) | **Grok** (xAI)                     |
| **Kimi** (Moonshot)                              | **Llama 4 Maverick** (Meta)        |
| **Qwen** (Alibaba)                               | **DeepSeek**                       |

Different models have different strengths. Some are more literary, some faster, some better at following exact instructions. If a chat grows longer than a given model's context window, that model is disabled for that thread until you trim it.

<Note>
  On **Basic**, Chat runs on a strong, fast default model chosen for you. **Model selection** is a Pro feature. See [Plans & pricing](/account/plans).
</Note>

## Different jobs, different models

Not everything uses the big frontier models. Quarkle matches the model to the job.

* **Chat & brainstorming.** Capable conversational models (your pick on Pro, a fast default on Basic).
* **The Review Room.** Models that reliably return **structured** feedback, run many at once per chapter.
* **Story Atlas upkeep.** A cheap, fast model that can read a diff and update notes, running quietly in the background.
* **Behind the scenes.** Small, fast models handle quick jobs like classifying your intent, condensing very long chats, and detecting a book's genre. An **embeddings** model powers semantic search over your manuscript and past chats.

## Providers

Models are reached through a few providers: primarily **OpenRouter**, with **AWS Bedrock** for some models, and dedicated providers for the embeddings that power search. This multi-provider setup is what lets Quarkle offer a wide model menu and stay up when any single provider has an outage.

<Tip>
  Want a specific new model added? Pro users can ask. Email **[team@quarkle.ai](mailto:team@quarkle.ai)**. We tend to add notable new models within days of release.
</Tip>

<Card title="Next: Inside Quarkle Chat" icon="arrow-right" horizontal href="/under-the-hood/inside-quarkle-chat">
  How Chat decides what to read.
</Card>