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

# Choosing an AI Model

> Pick the right AI model for your task — from quick edits to complex architecture decisions

Vybe supports multiple AI models from Anthropic, OpenAI, and Google. Each model has different strengths, speed characteristics, and credit costs. You can switch models at any time without losing your conversation history.

***

## Available models

| Model                  | Provider  | Reasoning | Best For                                                    |
| ---------------------- | --------- | --------- | ----------------------------------------------------------- |
| **Opus 4.5** (default) | Anthropic | Medium    | General app building, complex logic                         |
| **Opus 4.5 - High**    | Anthropic | High      | Complex architectural decisions, debugging difficult issues |
| **Sonnet 4.5**         | Anthropic | Standard  | Faster responses, straightforward changes                   |
| **Haiku 4.5**          | Anthropic | Standard  | Quick edits, simple questions                               |
| **GPT-5.1**            | OpenAI    | Medium    | Alternative perspective, specific strengths                 |
| **GPT-5.1 - High**     | OpenAI    | High      | Complex reasoning with GPT                                  |
| **Gemini 3 Pro**       | Google    | Standard  | Alternative option, different approach                      |

***

## How to switch models

Click the **model selector** in the chat panel. A dropdown shows all available models. Select the model you want, and your next message will use it.

* Your model selection persists across sessions until you change it
* Switching models does not clear your conversation history
* The AI picks up right where the previous model left off

***

## Understanding reasoning effort

Some models offer configurable reasoning effort levels. Higher reasoning means the model spends more time thinking through complex problems before responding.

| Reasoning Level | Behavior                                                     | Credit Cost |
| --------------- | ------------------------------------------------------------ | ----------- |
| **Standard**    | Normal processing — good for most tasks                      | Lowest      |
| **Medium**      | Extended reasoning — better for multi-step logic             | Moderate    |
| **High**        | Deep reasoning — best for complex architecture and debugging | Highest     |

Models with "High" in their name (like Opus 4.5 - High) use high reasoning effort by default. Standard models use their default reasoning level.

***

## When to use which model

<Accordion title="Opus 4.5 — General app building">
  The default model and the best all-around choice. Use Opus 4.5 for:

  * Building new features from scratch
  * Writing complex business logic
  * Creating multi-page applications
  * Working with integrations and databases
  * Most day-to-day app development

  This is the recommended model for the majority of your work.
</Accordion>

<Accordion title="Opus 4.5 - High — Complex problems">
  The same model with higher reasoning effort. Use this when:

  * Debugging issues that Opus 4.5 (standard) could not resolve
  * Planning complex application architecture
  * Working through intricate data transformations
  * Handling edge cases in business logic

  Switch to this when you need the AI to think harder about a problem.
</Accordion>

<Accordion title="Sonnet 4.5 — Faster iteration">
  A faster model that handles straightforward tasks efficiently. Use Sonnet 4.5 for:

  * Simple UI changes (colors, layout, text)
  * Adding straightforward CRUD features
  * Quick code modifications
  * Iterating rapidly on visual design

  Sonnet 4.5 responds faster and uses fewer credits, making it ideal for rapid iteration cycles.
</Accordion>

<Accordion title="Haiku 4.5 — Quick edits">
  The fastest and most cost-effective model. Use Haiku 4.5 for:

  * Minor text changes
  * Simple styling updates
  * Quick questions about your code (in Ask mode)
  * Small, well-defined edits

  Haiku 4.5 is best for tasks where speed matters more than depth.
</Accordion>

<Accordion title="GPT-5.1 — Alternative perspective">
  An OpenAI model that offers a different approach. Use GPT-5.1 when:

  * You want a second opinion on an approach
  * Opus has difficulty with a specific task
  * You prefer GPT-style outputs for certain tasks

  Available in standard and high reasoning variants.
</Accordion>

<Accordion title="Gemini 3 Pro — Another alternative">
  A Google model offering yet another approach. Use Gemini 3 Pro when:

  * You want to try a different model's take on a problem
  * Exploring alternatives for specific use cases
</Accordion>

***

## Cost implications

More powerful models and higher reasoning levels consume more credits per message. As a general rule:

* **Haiku 4.5** uses the fewest credits per interaction
* **Sonnet 4.5** uses moderate credits
* **Opus 4.5** uses more credits but delivers the best results for complex tasks
* **High reasoning variants** use the most credits

<Tip>
  A practical workflow: use Opus 4.5 for initial app creation and complex features, then switch to Sonnet 4.5 or Haiku 4.5 for quick refinements and small edits. This balances quality with cost.
</Tip>

***

## Model selection strategy

Here is a practical approach to model selection:

1. **Start with Opus 4.5** for building new features and complex logic
2. **Switch to Sonnet 4.5** when iterating on UI and making straightforward changes
3. **Use Haiku 4.5** for quick questions and minor edits
4. **Upgrade to Opus 4.5 - High** when stuck on a difficult bug or architectural decision
5. **Try GPT-5.1** if you want an alternative perspective on a problem

***

## What's next

* Learn how the editor works: [The Editor](/getting-started/interface)
* Understand the live preview: [Preview & Sandbox](/building-apps/preview-and-sandbox)
* Tips for communicating with the AI: [Prompting Tips](/tips/prompting)