LLM Models

Regal offers multiple leading Large Language Models (LLMs) to power your AI agents for phone and SMS conversations. Each is optimized for different use cases – so the best choice depends on your goals, call type, complexity of conversation, and cost tolerance.

We recommend starting with OpenAI GPT-4o mini which is Regal's default model. We then recommend progressively upgrading to e.g., 4.1 mini, then 4.1, then 4o (or switch to Claude models) only if you find that your AI agent isn’t performing the call as well as you’d like, and you're willing to pay more.

GPT-4o Mini

Best for short, simple cost-sensitive use cases (e.g., calls normally staffed out of low cost locale)

Examples:

• Confirmation
• Scheduling
• Collections calls
• Qualification calls with < 15 questions (and not much branching)

✅ Pros: Fastest model, cost-effective, great for straightforward tasks/conversations. Included in Regal base AI voice agent minute pricing
❌ Cons: Slightly weaker at reasoning, so if your agent has to do some math, for example, then don’t use this model or build a custom action that does the math and returns the result to your agent

GPT-4.1 Mini

Best for short, simple conversations where strict task instruction following is needed (improvisation is not valued)

Examples:

• Confirmation
• Scheduling
• Collections calls
• Qualification calls with < 25 questions

✅ Pros: Slightly more reliable and consistent at following task instructions and function calling, has updated knowledge through 2024
❌ Cons:Slightly worse at improvisation, slightly more expensive.

GPT 4.1 or GPT-4o

Best for longer, more complex calls or ones that require math/reasoning

We've found GPT 4.1 to be better at following prompt instructions and function calling; but we've found GPT 4o to generate better conversational quality/improvisation.

Examples:

• Long qualification calls with 10+ steps or conditional steps
• More open-ended support calls
• Calls with many different custom actions to invoke, especially if iterative custom action calling
• Calls that require complex reasoning, such as math (e.g., agent needs to multiply the number of rooms the customer is moving + the number of boxes to estimate a quote)

✅ Pros: Pros: Handles conditional logic, multiple functions, and nuanced conversation better.
❌ Cons:Cons: More expensive; adds a few hundred milliseconds of latency – but most customers won’t notice.

When to Consider Claude

If you’ve tried OpenAI GPT 4o Mini or 4.0 and they aren’t getting the performance you need – especially around memory, creativity, or edge case handling – consider:

• Claude 3.5 Haiku for fast, lightweight conversations
• Claude 3.7 Sonnet for longer, more complex conversations

Switching Between Models

When you switch from one model to another, you might find that slight prompt changes are needed. That's expected. For 4o mini for example, you might have to prompt in a more particular way whereas 4o is less sensitive to your exact syntax, and can more easily figure out what you mean.

So don't YOLO switch your model without running a few tests and making sure your agent is behaving the way you want.

Still not sure? Just reach out – we’re happy to help you tune your agent for the best performance.

Want Regal to offer a different LLM? Just reach out – we’re constantly evaluating new LLMs and happy to onboard any that demonstrate differentiated performance.

Want to bring your own LLM? Just reach out – we’ll help you incorporate it into Regal.