Skip to content
Main Site News Console

Retell Powers 100 Million Monthly Calls with OpenAI Models

· OpenAI Translated
OpenAILLM

October 27, 2026

GPT‑6 Astra Tops Retell Voice Agent Benchmark, as Codex Helps Power $1.6M in ARR per Employee

Figure 1: Retell logo set against a blue acoustic fabric background.

  • Company Size: Startup

  • Region: North America

  • Industry: Technology

  • Products Used: API, Codex

  • 100M+: Calls handled per month

  • 80%+: Customer resolution rate

  • 1T: OpenAI tokens consumed

  • $1.6M: Annual revenue run rate per employee

Retell AI is a voice AI platform for contact centers, helping businesses build agents that answer questions, perform actions, and seamlessly transfer callers to human representatives when needed. The platform now serves over 4,000 businesses, with monthly call volume surging from 30 million at the beginning of 2026 to more than 100 million in September.

As Retell scales to support enterprise customers with increasingly complex workflows, it relies on OpenAI models to ensure exceptional call quality and reliability. Today, 80% of call responses on the platform are powered by OpenAI models, driving customer resolution rates past 80%. Retell has also crossed the milestone of 1 trillion total tokens consumed across OpenAI models, underscoring the massive scale of its real-world deployments.

Retell’s engineering team uses Codex extensively across its entire development cycle—from exploring unfamiliar codebases and building new features to debugging issues—while deploying GPT‑6 Astra to test its front-end interfaces.

“We choose models the same way our customers choose vendors: based on real-world call performance, not demo showcases. OpenAI continues to earn its place as our top choice.”

— Bing Wu, Co-Founder and CEO of Retell AI

Rigorously Selecting OpenAI Models for Real-Time Voice

Enterprise voice agents must adhere strictly to company policies and maintain their defined scope throughout evolving conversations. To evaluate options, Retell built its proprietary vCX benchmark, comparing dozens of large language models against complex, long-context, multi-turn dialogues pulled from real production scenarios.

The benchmark focuses on measuring instruction following, factual grounding based on provided information, accurate tool calling, and natural conversational cadence. In Retell’s benchmark release on October 13, 2026, GPT‑6 Astra took the top spot in instruction following with an 89.7% score.

Figure 2: Bar chart showing five models ranked on Retell’s vCX instruction following benchmark, led by GPT-6 Astra at 89.7%.

Retell also shared evaluation data conducted prior to the release of GPT‑6 Astra: with reasoning enabled, GPT‑5.5 ranked first in both hallucination-free rate (88.0%) and tool-calling accuracy (88.3%), outperforming Gemini‑3.1‑Pro and Claude‑4.6. Even with reasoning disabled, it retained the lead in hallucination-free performance.

Retell notes that enterprise customers who choose OpenAI models consistently expand their usage over time. Customers can also access the OpenAI Realtime model family via the Realtime API and leverage OpenAI’s text-to-speech (TTS) technology as their voice engine. Retell rolls out day-one support whenever OpenAI releases a new model, continuously updating benchmark evaluations as capabilities advance.

Resolving Over 80% of Customer Inquiries with Agents

Contact center teams use Retell to define their agents’ business context, goals, and rules, while developers configure complex workflows and backend integrations.

During live calls, Retell’s orchestration engine runs audio models, tracks conversational state, and precisely governs when the agent speaks, listens, or responds to user interruptions. OpenAI models—including GPT‑6.1 Sol, GPT‑6 Astra, and GPT‑5.5—generate responses and initiate tool calls informed by customer-defined instructions and ongoing dialogue context. Retell executes these calls through integrations with the customer’s internal systems, writing results directly back to their business records.

When human judgment is required, the voice agent briefs a human representative on the call context, answers any follow-up questions, and completes the transfer only after confirming the agent is ready. Once the call wraps up, Retell uses GPT‑5.6 Terra to analyze the transcript and extract disposition outcomes alongside key operational metrics.

Figure 3: Architecture diagram showing how Retell leverages OpenAI models to connect live audio, conversational state, context, tools, and customer systems during a live call.

This architecture enables Retell’s customers to resolve over 80% of customer inquiries autonomously. At consumer electronics brand Anker, agents handle frontline customer support across the US and UK, covering order tracking, warranty inquiries, and troubleshooting. Anker reports an 80.4% ticket resolution rate and a Net Promoter Score (NPS) of 63, noting that the agents’ tone, pacing, and warmth meet the same high standards expected of their human support teams.

At Pine Park Health, Retell agents manage appointment reminders and proactive scheduling outreach, logging outcomes directly into patient electronic medical records. This workflow has saved the equivalent of 2.1 full-time medical assistant roles, freeing staff to refocus on direct patient care. When reaching patients live, the scheduling agent achieves an appointment conversion rate of 55.7%.

Maintaining Low Latency Through Massive Traffic Surges

Daytime peak traffic on Retell’s platform can surge to more than 50 times its late-night lows. To sustain rapid conversational responsiveness during peak hours, Retell deploys shared, dedicated capacity in partnership with OpenAI. In Retell’s October 13, 2026 benchmark tests, GPT‑6 Astra and GPT‑6 Sol delivered a median first-sentence response latency of just 883 milliseconds, with GPT‑5.6 Luna leading the pack at 497 milliseconds.

OpenAI’s regionalized API endpoints allow developers to route traffic to specific geographic regions. This not only minimizes network latency for Retell, but also satisfies strict data residency and compliance requirements for clients in industries like healthcare and financial services.

Furthermore, Retell builds high-availability redundancy across every layer of its stack. It operates its own telephony stack…

#OpenAI#LLM#Codex#4Allapi.com

Published by the 4All API team

Original link:https://openai.com/index/retell-ai

Need a mainstream LLM API? 4All API gives you one key to call OpenAI, Anthropic, Google Gemini, Qwen, DeepSeek, and dozens more — at official-pass-through pricing, with enterprise-grade reliability, integrated in 5 minutes.

Sign up for the 4All API console →