Retell AI Powers 100M Calls Monthly with GPT-6

AI Voice Agents Reach New Heights

Retell's success with OpenAI models, particularly GPT-6 Astra and GPT-5.5, highlights a crucial trend: the maturation of LLMs for enterprise-grade applications requiring reliability, accuracy, and low latency. The benchmark results, especially GPT-6 Astra's top ranking in instruction following and GPT-5.5's performance in non-hallucination and tool-call correctness, are compelling. The article also effectively details how Codex accelerates development and debugging, a significant productivity booster. The $1.6M annual revenue run rate per employee is a striking metric, underscoring the efficiency gains. However, the article could delve deeper into the specific technical challenges of managing 100M monthly calls and the architectural choices Retell made beyond mentioning dedicated servers and regional API endpoints. More granular detail on how Retell ensures data privacy and security for sensitive customer data within contact center operations would also be beneficial, given the industries served.

Key Points

• Retell AI is successfully leveraging OpenAI's GPT-6 Astra and GPT-5.5 models to handle over 100 million monthly calls, achieving over 80% customer resolution rates.
• GPT-6 Astra leads in Retell's vCX benchmark for instruction following (89.7%), while GPT-5.5 excelled in non-hallucination (88.0%) and tool-call correctness (88.3%) in pre-launch evaluations.
• OpenAI's Codex significantly boosts engineering productivity by assisting in code research, feature building, and debugging, contributing to a high revenue run rate per employee ($1.6M).
• GPT-6 Astra is also used for automated testing of front-end interfaces, improving reliability and speed.
• Retell utilizes OpenAI's Realtime API and text-to-speech, making new models available on launch and rerunning benchmarks to stay updated.
• The platform ensures call responsiveness during peak demand using dedicated servers and regional API endpoints to minimize latency and meet data residency requirements.
• Retell's architecture includes redundancy at multiple layers, including telephony and model services, to ensure continuous operation.

---

📖 Source: Retell uses OpenAI models to help power 100M monthly calls