This guide is for dashboard users. Model selection is done inside the Kakiyo dashboard when configuring prompts. No code required.
The AI model you choose directly affects how intelligent and natural your conversations sound. A cheaper model saves credits per message, but poor-quality responses that prospects ignore waste more money than a better model that generates real replies.
Where to Change the AI Model
- In the sidebar, click Prompts.
- Click the prompt you want to edit.
- In the prompt editor, look for the AI Model dropdown.
- Select the model you want to use.
- Click Save.
The model applies to all campaigns that use this prompt. Changing the model does not affect ongoing conversations — only new messages generated after the change.
Available Model Families
| Family | Provider | Strengths |
|---|
| Claude (Sonnet, Haiku, Opus) | Anthropic | Excellent conversational AI, strong reasoning, natural tone, good at handling objections |
| GPT (GPT-4o, GPT-4o-mini, etc.) | OpenAI | Well-rounded performance, good general knowledge, reliable across conversation types |
| Gemini (Pro, Flash) | Google | Strong analytical capabilities, good conversation flow |
| Grok | xAI | Varied performance, worth testing for specific use cases |
Available models may change as new versions are released. Check the AI Model dropdown in your prompt editor for the current list. You can also see all available models via API.
Recommended Starting Point
Claude Sonnet 4.5 (Anthropic) is the recommended default model as of February 2026. It delivers an excellent balance of:
- Conversation quality and natural tone
- Objection handling capability
- Personalization based on prospect profiles
- Credit efficiency relative to output quality
Model performance evolves. What’s best today may not be best in six months. Re-test periodically and compare against newer models as they become available.
How to Test and Compare Models
Step 1: Pick Two Models to Compare
Choose your current model and one alternative you want to test.
Step 2: Test in the Sandbox
- Go to Prompts in the sidebar.
- Open the prompt you want to test.
- Set the model to Model A and save.
- Open the Sandbox for a campaign using this prompt.
- Run 3-5 conversations simulating different prospect types (interested, skeptical, busy).
- Note the quality of responses.
- Change the model to Model B and save.
- Run the same 3-5 conversations.
- Compare: which model produced more natural, relevant, and persuasive responses?
Step 3: Run a Live Comparison (Optional)
For more reliable data:
- Create two identical campaigns (same offering, same prompt text, same prospect profile).
- Set Campaign A to use Model A.
- Set Campaign B to use Model B.
- Import similar prospects into each (50-100 per campaign minimum).
- After 2 weeks, compare reply rates and conversation quality in Analytics.
Quality vs. Cost: How to Think About It
| Approach | When to use |
|---|
| Optimize for quality (use the best model) | High-value prospects, enterprise sales, complex products. Better conversations = more meetings = higher ROI even if credit cost is higher. |
| Optimize for cost (use a cheaper model) | High-volume, simpler outreach (event invites, content sharing). Messages are short and straightforward. |
| Balanced approach | Start with the recommended model, then test cheaper alternatives. Switch only if quality is comparable. |
Focus on cost per meeting booked, not cost per message. A model that costs 2x more per message but books 3x more meetings is the better choice.
Key Metrics to Track When Switching Models
After changing your AI model, monitor these metrics for 1-2 weeks:
| Metric | Where to find it | What to look for |
|---|
| Reply rate | Analytics > Campaign Stats | Did it go up or down? |
| Conversation quality | Inbox > Read conversations | Do responses sound natural? |
| Qualification rate | Analytics > Qualified Leads | Is the AI correctly qualifying? |
| Meeting booking rate | Analytics > Campaign Stats | More or fewer meetings? |
| Credit consumption | Billing > Credits | How much are you spending? |
