Retell AI scales to 100 million monthly calls using OpenAI’s models

Scaling Voice AI with OpenAI models

Retell AI’s voice platform now handles more than 100 million calls per month, up from 30 million earlier in 2026.

The growth accompanies service to over 4,000 businesses across North America.

OpenAI models generate responses for roughly 80 % of those interactions.

Customers using Retell report resolution rates that exceed 80 %.

Retell has consumed a total of one trillion tokens from OpenAI’s APIs.

The company cites an annual revenue run rate of $1.6 million per employee.

OpenAI’s Codex is employed by Retell engineers for code research, feature development, and debugging.

GPT‑6 Astra powers Retell’s front‑end testing and benchmark evaluations.

“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

Benchmark results and customer impact

Retell created a vCX benchmark that pits dozens of language models against real, multi‑turn call scenarios.

The benchmark measures instruction adherence, grounding, tool‑call accuracy, and conversational naturalness.

In the October 13 2026 benchmark, GPT‑6 Astra achieved the highest instruction‑following score at 89.7 %.

Prior to Astra’s launch, GPT‑5.5 led on non‑hallucination (88.0 %) and tool‑call correctness (88.3 %) with reasoning enabled.

GPT‑5.5 also topped the non‑hallucination metric when reasoning was disabled.

Both models outperformed Gemini‑3.1‑Pro and Claude‑4.6 on those measures.

Retell makes new OpenAI models available the day they are released and re‑runs its benchmark after each update.

Customers can also access OpenAI Realtime models via the Realtime API and OpenAI text‑to‑speech as a voice engine.

Technical workflow behind the calls

During a call, Retell’s harness runs audio models, tracks conversation state, and decides when the agent should speak, listen, or handle interruptions.

OpenAI models such as GPT‑6.1 Sol, GPT‑6 Astra, and GPT‑5.5 use the caller’s instructions and context to generate replies and request tool calls.

Retell executes those tool calls through integrations with the client’s own systems and can write outcomes back to a customer record.

If human judgment is required, the voice agent briefs a live representative with relevant context before transferring the call.

After the call ends, Retell employs GPT‑5.6 Terra to analyze the transcript, extract outcomes, and capture additional data.

At electronics retailer Anker, the Retell agent handles Tier‑1 post‑sales support in the United States and United Kingdom.

Anker reported an 80.4 % case‑resolution rate and a Net Promoter Score of 63 while using the platform.

The combined performance metrics illustrate that Retell’s integration of OpenAI models can sustain high‑volume, high‑resolution contact‑center operations.

Why This Matters: Retell’s ability to process over 100 million AI‑driven calls each month while maintaining strong resolution rates demonstrates that large‑scale enterprise voice AI is now operationally viable, opening a path for broader adoption in contact‑center automation.