Best Practices
Model Selection Guide
Learn how to choose the right AI model for your camelAI iframe implementation
camelAI Legacy Product — This documentation covers camelAI's embedded analytics offering, which is no longer being actively developed. We are migrating existing customers to the new camelAI platform. For the current product, visit camelAI.
Selecting the appropriate model is crucial for optimizing your users' experience with camelAI. Different models offer unique strengths in terms of intelligence, speed, cost, and interaction style. This guide will help you choose the model that best fits your use case.
How Model Selection Works
When you create an iframe through the camelAI API, you specify which model to use. This model choice is locked for that specific chat conversation - ensuring consistency throughout the user's interaction. If users access previous conversations, camelAI uses the model that was originally selected for that chat. If the provider has shut that model down, the chat continues on its successor (see Older Model IDs Are Aliases).
Available Models
We currently support seven models across two providers. Pass the model ID as model when you create an iframe or call Ask Camel.
Model Comparison
| Model | Model ID | Provider | Best For | Speed | Cost | Intelligence |
|---|---|---|---|---|---|---|
| GPT-6.1 Sol (Default) | gpt-6.1-sol | OpenAI | Most workloads: complex queries, multi-step analysis | Medium | Medium | Very High |
| GPT-6 Astra | gpt-6-astra | OpenAI | The hardest questions, where accuracy matters most | Slow | Highest | Highest |
| GPT-6 Luna | gpt-6-luna | OpenAI | Simple data, high-volume or cost-conscious deployments | Very Fast | Lowest | High |
| Claude Sonnet 5.5 | sonnet-5.5 | Anthropic | In-depth research, comprehensive analysis | Medium | Medium | Very High |
| Claude Opus 5.5 | opus-5.5 | Anthropic | Long, open-ended investigations | Slow | High | Highest |
| Claude Haiku 5.5 | haiku-5.5 | Anthropic | Fast, straightforward questions | Very Fast | Lowest | High |
| Claude Fable 5.1 | fable-5.1 | Anthropic | The most demanding analysis, regardless of cost | Slow | Highest | Highest |
Model Characteristics
Our recommended default, with strong reasoning at a moderate cost.
Use GPT-6.1 Sol when:
- Your data requires complex joins or calculations
- You want one model that handles most questions well
Older Model IDs Are Aliases
Model IDs from earlier versions of this guide still work. Each one is an alias for the current model in the same family and tier. Aliases apply to new conversations: a new iframe chat or Ask Camel request that names an older ID runs on its successor.
| Older ID | Runs on |
|---|---|
gpt-5 | gpt-6.1-sol |
gpt-5-codex | gpt-6.1-sol |
o3 | gpt-6.1-sol |
gpt-4.1 | gpt-6.1-sol |
o4-mini | gpt-6-luna |
sonnet-4 | sonnet-5.5 |
sonnet-4.5 | sonnet-5.5 |
haiku-4.5 | haiku-5.5 |
New chats record the model that actually ran, so a request for o3 creates a chat on gpt-6.1-sol. An ID that is neither a current model nor an alias returns a 400 error.
Existing conversations keep the model they were created with. The exception is a model the provider has shut down: chats on sonnet-4, gpt-5-codex or o4-mini continue on sonnet-5.5, gpt-6.1-sol and gpt-6-luna.
Best Practices
Test Before Production
We strongly recommend experimenting with different models using your actual data before deploying to production. Model performance can vary significantly based on:
- Your data structure complexity
- Typical query patterns
- User expectations
- Performance requirements
Updates and Support
We continuously evaluate and update our model offerings to provide the best possible experience. This guide will be updated as new models become available or existing models are improved.
If you don't see a model that meets your requirements, we'd love to hear from you. Contact us to request additional model support.