GPT-5 is now live on Retell.

We've added GPT-5 family to our platform — the most advanced language model yet. This means smarter conversations and, in many cases, lower costs for your agents.

What does GPT 5 mean for AI voice agents?

GPT-5 is a reasoning model, and as the comparisons on OpenAI’s site show, it delivers significantly higher intelligence than previous models. Its advanced reasoning allows for:

• Deeper understanding of complex instructions
• Stronger context retention over multi-turn conversations
• More accurate and nuanced responses across varied scenarios

Reasoning models are typically slower than lighter models, which can impact real-time applications where sub-second response is critical.

To address this, we’ve integrated GPT-5 Minimal into our voice agent option. This variant preserves some of GPT-5’s reasoning capabilities while being tuned for faster response times. While it doesn’t reach the full intelligence level of the standard GPT-5 model, it offers a balance between improved reasoning and acceptable latency for real-time use.

This is the first time we are deploying a reasoning model in a real-time agent. We are actively experimenting to see how far we can push optimization without sacrificing quality. For now, GPT-5 Minimal is used in the backend to balance speed and intelligence, and we’ll continue refining the approach as we learn from live usage.

Which Model Should You Choose?

1. For high-intelligence, complex workflows

(Examples: advanced customer support, appointment scheduling, heavy function calling)

Recommendation:

GPT-5 offers slightly higher intelligence compared to GPT-4.1 and GPT-4o, but at the cost of slower response times due to its reasoning capabilities. For most real-time voice agent use cases where speed is critical, GPT-5mini or GPT 4.1 will be a better choices. GPT-5 becomes more viable if future improvements reduce its latency.‍

‍

2. For lightweight, high-speed tasks

(Examples: lead qualification, survey intake)

‍

Recommendation:

For small tasks that still benefit from reasoning, GPT-5 Mini is a strong choice. But if your workflow is speed-sensitive and doesn’t require deep reasoning, stick with GPT-4.1 Mini. If cost is the top priority, GPT-5 Nano offers the cheapest price while still maintaining high intelligence.

‍

In summary

• GPT-5: Intelligence unlock if latency improves‍
• GPT-5 Mini: Lower cost, strong intelligence, good balance for real-time agents‍
• GPT-5 Nano: Best for cost-sensitive businesses without needing a reasoning or latency boost
  

In testing, GPT-5 models showed slightly higher latency, possibly due to high early traffic, but cost efficiency and intelligence gains make them worth exploring. We’ll continue monitoring performance.

‍

Looking Ahead: What GPT-5 Could Unlock

GPT-5 brings us closer to agents that can:

• Follow complex instructions more reliably, enabling more complicated multi-step workflows
• Perform multi-step reasoning seamlessly during live calls
• Adapt conversation style dynamically to user emotion and intent
• Create and call tools using natural language (not just JSON) *Although it is already supported in other models, so it’s not unique to GPT-5.

We’ll continue to experiment and roll out these capabilities as they mature.

‍