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

# LLM Models Overview

> Understanding supported models and providers

Endprompt supports multiple LLM providers, giving you flexibility to choose the best model for each use case.

## Supported Providers

<CardGroup cols={3}>
  <Card title="OpenAI" icon="robot">
    GPT-4o, GPT-3.5-turbo, gpt-image-1
  </Card>

  <Card title="Anthropic" icon="message">
    Claude 3 Opus, Sonnet, Haiku
  </Card>

  <Card title="Google" icon="google">
    Gemini 2.0 Flash, Gemini 2.5 Pro
  </Card>
</CardGroup>

## Model Comparison

| Model                | Provider  | Speed     | Quality   | Cost      | Vision | Image Gen | Best For                          |
| -------------------- | --------- | --------- | --------- | --------- | ------ | --------- | --------------------------------- |
| **GPT-4o**           | OpenAI    | Fast      | Excellent | Medium    | ✅      | ❌         | General purpose, vision, balanced |
| **GPT-4**            | OpenAI    | Medium    | Excellent | High      | ✅      | ❌         | Complex reasoning                 |
| **GPT-3.5-turbo**    | OpenAI    | Very Fast | Good      | Low       | ❌      | ❌         | Simple tasks, high volume         |
| **gpt-image-1**      | OpenAI    | Medium    | Excellent | Per-image | ❌      | ✅         | Image generation and editing      |
| **Claude 3 Opus**    | Anthropic | Medium    | Excellent | High      | ✅      | ❌         | Nuanced analysis, long context    |
| **Claude 3 Sonnet**  | Anthropic | Fast      | Very Good | Medium    | ✅      | ❌         | Balanced performance              |
| **Claude 3 Haiku**   | Anthropic | Very Fast | Good      | Low       | ✅      | ❌         | Fast responses, simple tasks      |
| **Gemini 2.0 Flash** | Google    | Very Fast | Good      | Low       | ✅      | ❌         | Fast multimodal tasks             |
| **Gemini 2.5 Pro**   | Google    | Medium    | Excellent | Medium    | ✅      | ❌         | Complex reasoning, long context   |

## Choosing a Model

### For Quality-Critical Tasks

* **GPT-4o** or **Claude 3 Opus**
* Complex analysis, nuanced responses
* Higher cost, worth it for important outputs

### For High-Volume Tasks

* **GPT-3.5-turbo** or **Claude 3 Haiku**
* Simple classification, extraction
* Low cost, high throughput

### For Long Documents

* **Claude 3** models
* Support up to 200K tokens context
* Ideal for document analysis

### For Image Tasks

* **Vision** (image inputs): GPT-4o, Claude 3 models, Gemini — analyze images alongside text prompts
* **Image Generation** (image outputs): gpt-image-1 — generate or edit images from text descriptions
* **Image Editing**: Use gpt-image-1 with both image inputs and outputs to edit existing images

## Model Capabilities

### Context Window

How much input each model can handle:

| Model           | Max Input Tokens |
| --------------- | ---------------- |
| GPT-4o          | 128,000          |
| GPT-4           | 128,000          |
| GPT-3.5-turbo   | 16,385           |
| Claude 3 Opus   | 200,000          |
| Claude 3 Sonnet | 200,000          |
| Claude 3 Haiku  | 200,000          |

<Tip>
  For very long documents, Claude models offer the largest context windows.
</Tip>

### JSON Mode

Some models support native JSON output mode:

| Model           | JSON Mode     |
| --------------- | ------------- |
| GPT-4o          | ✅ Supported   |
| GPT-4           | ✅ Supported   |
| GPT-3.5-turbo   | ✅ Supported   |
| Claude 3 models | Via prompting |

<Note>
  JSON mode is enabled automatically when you define an output schema.
</Note>

## Model Selection in Prompts

When creating or editing a prompt:

1. Click the **Model** dropdown
2. Select your preferred model
3. The model is saved with the prompt

Different prompts on the same endpoint can use different models—useful for A/B testing.

## Cost Considerations

LLM costs are based on tokens:

* **Input tokens**: Your prompt text (including rendered template)
* **Output tokens**: The model's response

| Cost Factor     | Impact                                              |
| --------------- | --------------------------------------------------- |
| Model choice    | Higher-end models cost more per token               |
| Prompt length   | Longer prompts = more input tokens                  |
| Response length | Higher max\_tokens = potentially more output tokens |
| Request volume  | More requests = more total cost                     |

<Tip>
  Start with a capable model (GPT-4o), then optimize to cheaper models for tasks where quality remains acceptable.
</Tip>

## Model-Specific Tips

<Tabs>
  <Tab title="OpenAI">
    **GPT-4o** is the recommended default:

    * Best balance of speed, quality, and cost
    * Multimodal capabilities (can process images)
    * Reliable JSON output

    **Tips:**

    * Use temperature 0-0.3 for deterministic tasks
    * Enable JSON mode for structured output
    * GPT-3.5-turbo is 10x cheaper for simple tasks
  </Tab>

  <Tab title="Anthropic">
    **Claude 3 Sonnet** is the recommended default:

    * Excellent reasoning capabilities
    * Large context window (200K tokens)
    * Good balance of speed and quality
    * Vision support for image analysis

    **Tips:**

    * Great for document analysis and summarization
    * More nuanced at following complex instructions
    * Opus for the highest quality, Haiku for speed
  </Tab>

  <Tab title="Google">
    **Gemini 2.0 Flash** for fast multimodal tasks:

    * Built-in vision support
    * Competitive pricing
    * Large context window (1M tokens)

    **Tips:**

    * Excellent at image understanding tasks
    * Gemini 2.5 Pro for complex reasoning
    * Great balance of speed and capability
  </Tab>
</Tabs>

## Adding Provider API Keys

Your tenant needs API keys configured for each provider you want to use:

1. Go to **LLM API Keys** in the sidebar (under Configuration)
2. Add your OpenAI and/or Anthropic API keys
3. Keys are encrypted and stored securely

<Warning>
  Without a valid API key for a provider, prompts using that provider's models will fail.
</Warning>

## Next Steps

<CardGroup cols={2}>
  <Card title="Model Settings" icon="sliders" href="/llm-models/model-settings">
    Configure temperature, tokens, and more
  </Card>

  <Card title="API Authentication" icon="key" href="/api/authentication">
    Set up keys to call your endpoints
  </Card>
</CardGroup>