Skip to main content

​
What are AI Models?

Availability: Free and Starter plans can use GPT-5.6 Luna and GPT-5.4 Nano. Advanced and Business plans have access to the full model selection.
The AI model determines how your Agent interprets questions, follows Instructions and generates responses. For most AI Agents, we recommend GPT-5.6 Luna. It performs especially well with complex Instructions, Actions, conditions and multi-step workflows. GPT-5.4 mini is a strong alternative when you want more control over your Agent’s personality and tone of voice, while still maintaining strong overall performance.

​
Which model do we recommend?

OpenAI positions GPT-5.6 Luna for efficient, high-volume workloads and gives the model a 1.05M context window. See OpenAI’s GPT-5.6 Luna documentation. For more information about GPT-5.4 mini, see OpenAI’s GPT-5.4 mini documentation. If you are unsure which model to choose, start with GPT-5.6 Luna.

​
What about older AI models?

Older models are also still available for existing Agents. These models can still handle common tasks such as answering FAQs, responding to product or service questions, following Instructions, and using Sources. If your Agent is already performing well on an older model, you do not necessarily need to switch immediately. However, newer models generally offer improvements in areas such as reasoning, instruction following, speed, and efficiency. For new Agents, we therefore recommend GPT-5.6 Luna for most use cases. When an older model is scheduled for retirement, we will notify affected customers in advance and provide a deadline for switching models. For step-by-step instructions, see Migrate your AI model.

​
How do I change my AI model?

You can change the model without creating a new Agent.
  1. Open your Agent.
  2. Open the AI model setting.
  3. Select the model you want to use.
  4. Save your changes.
  5. Test your Agent in the Playground.
Your Agent Settings, Instructions, and Sources remain in place when you change the model.

​
Test your Agent after changing models

Different models can interpret the same Instructions and customer questions differently. Always test your Agent after changing models before relying on the new model in customer conversations. Use real customer questions from your Inbox to test:
  • Frequently asked questions
  • Important Instructions
  • Complex conversation flows
  • Actions, where relevant
  • Human handovers
  • Tone of voice
  • Language behaviour, if relevant
If important behaviour changes after switching models, review the relevant Agent Settings and Instructions before adding additional prompts. When comparing models, use the same conversations and scenarios so you can clearly evaluate the difference in behaviour.

​
Still unsure which model to choose?

Start with GPT-5.6 Luna and test your most important customer scenarios in the Playground. If you want more control over personality and tone of voice, compare those same conversations with GPT-5.4 mini.