Compare Models

Use Compare Models when you want to choose a model using your own knowledge, instructions, and customer questions. The page sends one message to every active panel, making differences easier to see than testing each model separately.

Every generated answer uses AI credits according to that model’s credit cost.

Start a comparison

1. Open a chatbot and select Compare Models.
2. Keep the two starting panels or select Add instance to create a third.
3. Choose a model and temperature for each panel.
4. Enter one realistic customer question in the shared message box.
5. Send the message and wait for every panel to finish.

All panels use the chatbot’s currently saved instructions. They also search the same trained Knowledge Base, so the comparison focuses on the selected model and temperature.

Configure each panel

Select the model name in a panel to start from the Recommended tab or open All models for the complete provider catalog. Use the search field to find a group, model, or provider. Restricted models appear faded and remain unavailable until the account has access to them.

Open the settings control to change Temperature from 0 to 2:

• Use a lower value for consistent, factual support answers.
• Use a middle value when you want natural wording with moderate variation.
• Use a higher value only when creativity matters more than predictability.

Changing a panel’s model or temperature does not change the chatbot’s public configuration. Comparison settings are kept in this browser for the current chatbot so you can continue testing later.

Manage comparison panels

Open a panel’s three-dot menu to move it left, move it right, or remove it. You can compare up to three panels and must keep at least one.

Select Reset to clear the comparison and return to two fresh panels. Reset also clears the comparison settings saved in this browser.

What to evaluate

Do not choose a model from one impressive answer. Use a small test set that covers:

• A common question with a direct answer
• A question whose answer is spread across related source content
• A conversational rewording of the same question
• A follow-up question that depends on the previous turn
• A question the Knowledge Base cannot answer
• A request that should trigger a plugin

Review each answer for accuracy, use of the available source information, instruction-following, clarity, tone, and whether it admits when information is missing. Consider response speed and credit cost after the answer quality is acceptable.

Apply your choice

Compare Models is a testing workspace. It does not publish the model or temperature selected in a panel.

When you have chosen a setup:

1. Open Playground.
2. Select the preferred model and temperature in the AI tab.
3. Select Save changes.
4. Test the public chatbot once more.

Troubleshooting

A model cannot be selected

Model availability depends on the current plan. Choose an available model or review the account plan under Billing.

The panels give very similar answers

That can be a good sign when the source contains one clear answer. Try more demanding questions, different temperatures, or examples that test boundaries and follow-up context. Do not regenerate solely to force visible differences.

The answers are all incorrect

The issue is probably the shared source content or saved instructions rather than every model. Confirm the answer exists in a Trained source, remove contradictions, and retest in Playground before comparing again.

Next step

Open Chat Logs to see how the chosen chatbot performs in real conversations.