Retell AI Leverages OpenAI Models to Achieve 100 Million Monthly Calls

Retell AI is a voice AI platform designed for contact center operations, enabling businesses to create agents capable of answering questions and connecting callers with human representatives. The platform now supports over 4,000 businesses, with monthly call volume increasing from 30 million in early 2026 to over 100 million by September 2026.

As Retell expands its services, it relies on OpenAI models to maintain high call quality and reliability. Currently, OpenAI models generate responses for 80% of calls on Retell, achieving resolution rates exceeding 80%. The platform has surpassed 1 trillion tokens consumed across OpenAI models, indicating significant adoption.

Retell’s engineers utilize Codex throughout the development process, from researching unfamiliar code to building features and debugging issues, while GPT-6 Astra is employed to test front-end interfaces.

“We choose models the way our customers choose vendors: on real calls, not demos. OpenAI keeps earning its spot.” —Bing Wu, Co-founder and CEO, Retell AI

Evaluating OpenAI Models for Live Call Demands

Enterprise voice agents must adhere to company policies and remain within their designated roles. Retell developed its vCX benchmark to evaluate various models using multi-turn conversations derived from real scenarios.

This benchmark assesses models’ effectiveness in following instructions, remaining grounded in provided information, and utilizing natural conversational language. In Retell’s October 13, 2026 benchmark, GPT-6 Astra achieved the highest score for instruction following at 89.7%.

Prior to GPT-6 Astra’s launch, evaluations indicated that GPT-5.5 ranked highest for non-hallucination rate (88.0%) and tool-call correctness (88.3%) with reasoning enabled, outperforming Gemini-3.1-Pro and Claude-4.6.

New customers opting for OpenAI models continue to use them, with access to OpenAI Realtime models via the Realtime API and OpenAI text-to-speech as a voice engine. Retell makes new OpenAI models available on launch day and re-evaluates its benchmark as models improve.

High Resolution Rates with Voice Agents

Contact center teams utilize Retell to define their agents’ business context, goals, and rules. During calls, Retell’s system manages audio models and controls when the agent speaks or responds to interruptions. OpenAI models generate responses based on customer instructions and conversation context.

When human judgment is required, the voice agent briefs the representative with context and confirms readiness before connecting the customer. After calls, Retell employs GPT-5.6 Terra to analyze transcripts and extract outcomes.

With this approach, Retell reports customer resolution rates above 80%. For example, electronics company Anker utilizes agents for Tier-1 post-sales support, achieving an 80.4% case resolution rate and a Net Promoter Score (NPS) of 63.

At Pine Park Health, Retell agents manage appointment reminders, recording outcomes in patients’ medical charts, freeing up medical-assistant time equivalent to 2.1 full-time roles.

Maintaining Responsiveness Amid High Call Volume

Retell’s platform experiences daytime traffic that can exceed 50 times its overnight low. To ensure responsiveness during peak demand, Retell employs dedicated servers with OpenAI, shared across its customer base. In the October 13, 2026 benchmark, GPT-6 Astra and GPT-6 Sol recorded median first-sentence latency of 883 milliseconds, while GPT-5.6 Luna achieved the lowest latency at 497 milliseconds.

OpenAI’s regional API endpoints enable developers to direct traffic to specific geographic areas, helping Retell minimize network travel time and comply with data residency requirements.

Retell incorporates redundancy at every service layer, operating its own telephony stack with backups ready to ensure continuity during service failures.

Utilizing Codex for Development and Debugging

As Retell’s enterprise growth leads to increased requests for custom workflows, engineers often navigate unfamiliar code segments. They provide Codex with high-level requirements, allowing it to survey the code and suggest necessary changes.

During implementation, Codex identifies details engineers may have overlooked. For debugging, it cross-references logs and database records with the code. Retell’s engineers also use Codex to assist customers building on OpenAI models.

To test modifications to its web interface, Retell relies on GPT-6 Astra’s capabilities, which interact with the interface and report any issues found.

While Retell continues to hire, its engineering team has not needed to expand at the same rate as the business, reporting an annual revenue run rate of $1.6 million per employee.

“By using Codex, the engineering team has a virtual teammate that can pair with them on any kind of difficult task.” —Bing Wu, Co-founder and CEO, Retell AI

Empowering Customer Expertise

Retell aims to empower businesses to apply their expertise directly, utilizing advanced voice AI. Looking ahead, Retell plans to expand quality assurance, support additional languages, and develop knowledge base auto-learning.

As OpenAI models evolve, Retell anticipates that its customers will provide exceptional service to millions of callers, tailored to their specific needs.

 

Source: OpenAI News