For real-time phone voice agents, is GPT-4o or GPT-4.1 better?

Hi miarayhan, For real-time phone voice agents, the choice between GPT-4o and GPT-4.1 depends on the specific needs of your application. - GPT-4o is specially designed for real-time applications like phone agents and includes features tailored for speech-to-speech models as described in [OpenAI Realtime](https://docs.vapi.ai/openai-realtime). - GPT-4.1 models may offer improved performance or new features over previous versions, though they are generally more focused on better text processing capabilities and might provide more stable and precise outputs for voice synthesis and understanding, as detailed in the [April 15, 2025 changelog](https://docs.vapi.ai/changelog/2025/4/15). If you're prioritizing real-time speech capabilities, you might lean towards GPT-4o. However, if your focus is on newer enhancements and improved text processing, GPT-4.1 might be the way to go. Please do let us know if you need further assistance about this and we will gladly investigate further.

Thanks a lot. I will do some tests on both and I will share my results soon.

Thanks. We will await for your results!

Thanks, In our experiment, we found 4o to perform best. It's almost all the time stable and unlike 4.1. The tool calling is also better with 4o.

Great to hear! We typically use 4o with our test assistant as well. If you happen to find any issues with it at any point, please feel free to reach out

Aer you still using 4o for live ?

Yes, we typically use 4o or 4o-mini, but really depends on your use-case and how fast you want the latency vs. quality of responses