Industry

Retell AI Powers 100M Monthly Calls With OpenAI Models, Achieving 80%+ Resolution Rates

Voice AI platform Retell scales to 100M monthly calls using OpenAI models for 80% of interactions, with customers resolving over 80% of requests without human intervention.

Last verified:

Retell AI Scales to 100M Monthly Calls on OpenAI Model Infrastructure

Retell AI, a voice AI platform for enterprise contact centers, has reached a milestone of over 100 million calls processed monthly as of September 2026—a more than threefold increase from 30 million in early 2026. According to the OpenAI Blog, the platform now serves more than 4,000 businesses and has consumed over 1 trillion tokens across OpenAI models, underscoring the scale at which the company relies on OpenAI’s inference infrastructure.

OpenAI models power the vast majority of Retell’s inbound interaction logic. The OpenAI Blog reports that OpenAI models generate responses for 80% of calls on Retell’s platform, with customers achieving resolution rates above 80%—meaning the automated agent handles the majority of customer requests without requiring human escalation.

GPT-6 Astra Leads Retell’s Enterprise Voice Benchmark

To evaluate model performance under real-world contact center constraints, Retell developed the vCX benchmark, which tests multi-turn conversations drawn from production scenarios. The benchmark assesses instruction-following, hallucination resistance, tool-call accuracy, and conversational naturalness—capabilities critical for agents that must stay grounded in company policies and customer context.

In Retell’s October 13, 2026 benchmark results, GPT-6 Astra ranked first for instruction-following accuracy at 89.7%. According to the OpenAI Blog, earlier evaluations conducted before GPT-6 Astra’s launch showed GPT-5.5 leading on non-hallucination rate (88.0%) and tool-call correctness (88.3%) with reasoning enabled, outperforming both Gemini 3.1 Pro and Claude 4.6 on both dimensions.

Retell’s Integration Strategy and Developer Workflow

Retell’s architecture layers OpenAI models into a broader voice orchestration harness. During a call, the system manages audio transcription, conversation state tracking, and agent interruptibility. OpenAI models—including GPT-6.1 Sol, GPT-6 Astra, and GPT-5.5—interpret customer instructions and conversation context to generate responses and invoke tool calls. Retell then executes those integrations against customer backend systems (CRM, ticketing, etc.) and can persist outcomes to customer records.

Retell’s engineering team also uses OpenAI’s development tools. According to the OpenAI Blog, engineers leverage Codex throughout development for code research, feature engineering, and debugging, and use GPT-6 Astra to test front-end interfaces.

Retell makes new OpenAI models available on launch day and re-runs its vCX benchmark as model capabilities improve. The platform also offers access to OpenAI’s Realtime API and text-to-speech service as alternative voice engines, giving customers optionality while maintaining OpenAI as the primary agentic backbone.

Why This Matters

Retell’s 3x scaling in call volume within nine months signals strong enterprise demand for AI-powered contact center automation. The company’s reliance on OpenAI models for 80% of calls—and its stated practice of choosing models “on real calls, not demos”—validates OpenAI’s competitive position in agentic, low-latency inference where instruction-following and tool-use accuracy are non-negotiable.

For contact center buyers evaluating voice AI platforms, Retell’s public benchmarking methodology (vCX) offers a rare, production-grounded comparison standard. The 80%+ resolution rate indicates that GPT-6 Astra and GPT-5.5 have crossed a maturity threshold for real-money use cases. Retell’s decision to re-benchmark quarterly and expose model performance data publicly also raises the competitive bar for vendors—open benchmarks tend to accelerate adoption of top performers and force laggards to improve.

Frequently Asked Questions

What percentage of Retell's calls use OpenAI models?

According to the OpenAI Blog, OpenAI models generate responses for 80% of calls on Retell's platform.

How much has Retell's call volume grown in 2026?

Call volume grew from 30 million in early 2026 to over 100 million by September 2026—more than a 3x increase.

Which OpenAI model performs best on Retell's vCX benchmark?

GPT-6 Astra ranked first for instruction-following at 89.7% on Retell's October 13, 2026 vCX benchmark.

What is Retell's customer resolution rate?

Customers using Retell achieve resolution rates above 80%, meaning the voice agent handles over 80% of requests without escalating to a human representative.

#voice-ai #openai #contact-centers #enterprise-ai #real-time-models