Retell Scales Voice AI to 100M Calls Using OpenAI Models

Introduction

Voice AI platform Retell has reached a significant milestone, now serving over 100 million monthly calls across more than 4,000 businesses. The company attributes its ability to scale while maintaining call quality and responsiveness to its deep integration with OpenAI models. As enterprises demand more complex and reliable voice agents, Retell’s architecture and development practices offer insights into how large-scale conversational AI can be deployed effectively in production environments.

Scaling Voice Agents with OpenAI Models

Retell’s platform generates responses for 80% of live calls using OpenAI models, with customers achieving resolution rates above 80%. The company has consumed over 1 trillion tokens across these models, underscoring the breadth of its deployment. To ensure consistent performance under load, Retell uses dedicated servers shared across customers and leverages OpenAI’s regional API endpoints to reduce latency and meet data residency requirements in regulated industries like healthcare and finance. During peak daytime traffic, which can exceed 50 times overnight lows, Retell maintains responsiveness through infrastructure redundancy and optimized routing.

Benchmarking Model Performance for Enterprise Use

To evaluate model suitability for enterprise voice agents, Retell developed its vCX benchmark, which tests models on challenging multi-turn conversations drawn from real production scenarios. In its October 13, 2026 benchmark, GPT-6 Astra ranked first for instruction following at 89.7%, while GPT-6 Astra and GPT-6 Sol each recorded median first-sentence latency of 883 milliseconds. Earlier evaluations showed GPT-5.5 leading in non-hallucination rate (88.0%) and tool-call correctness (88.3%) with reasoning enabled. These benchmarks help Retell and its customers choose models that balance accuracy, speed, and compliance with business policies.

Accelerating Development with Codex

Beyond runtime performance, Retell’s engineering team uses Codex throughout the development lifecycle. Engineers provide high-level requirements to Codex, which then surveys the codebase, checks it against real data, and identifies necessary changes—often faster than manual review. During implementation, Codex helps catch overlooked details, and for debugging, it cross-references logs and database records with code to investigate issues. This workflow supports Retell’s need to build custom integrations and workflows for enterprise clients operating in unfamiliar code paths.

Real-World Impact Across Industries

Retell’s voice agents are delivering measurable outcomes across sectors. At Anker, agents handle Tier-1 post-sales support in the US and UK, achieving an 80.4% case resolution rate and a Net Promoter Score of 63, with tone and pacing meeting human-agent standards. At Pine Park Health, agents manage appointment reminders and proactive scheduling, recovering the equivalent of 2.1 full-time medical assistant roles and achieving a 55.7% booking rate when reaching patients. These results highlight how voice AI can reduce operational burden while maintaining service quality.

Conclusion

Retell’s growth to over 100 million monthly calls powered by OpenAI models reflects the increasing viability of voice AI in enterprise contact centers. With strong resolution rates, scalable infrastructure, and efficient development workflows, the company sets a benchmark for others looking to deploy AI agents at scale. As demand for intelligent automation continues to rise, platforms like Retell are well-positioned to integrate AI automation and DevSecOps practices into cloud-native contact center solutions.