TL;DR
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
Retell says its voice-agent platform handled more than 100 million calls in September 2026, up from 30 million in early 2026, and now serves more than 4,000 businesses. The company says OpenAI models generate responses for 80% of its calls and that customers achieve resolution rates above 80%; those figures come from OpenAI’s account of Retell’s deployment.
Retell AI says its voice-agent platform handled more than 100 million calls in September 2026, up from 30 million in early 2026, as it expanded to more than 4,000 businesses. In a report published by OpenAI on October 13, the companies said OpenAI models generate responses for 80% of Retell calls and that Retell customers record resolution rates above 80%.
The figures are presented in an OpenAI account of its work with Retell, rather than in an independently audited performance report. OpenAI says Retell has consumed more than 1 trillion tokens across its models. Retell’s platform is designed for contact-center operations: agents answer questions, take actions through business-system integrations and hand callers to a human when needed.
Retell’s software manages the live-call process, including tracking the conversation and handling when an agent listens, responds or deals with an interruption. OpenAI models use customer-provided instructions and conversation context to draft replies or request tool calls; Retell then runs those calls through customer integrations. The company also uses a model to analyze transcripts and extract call outcomes after conversations end.
OpenAI reports that customers can use models including GPT-6.1 Sol, GPT-6 Astra and GPT-5.5 for responses, while GPT-5.6 Terra is used for post-call transcript analysis. The report also describes access to OpenAI’s Realtime API and text-to-speech. Retell says it tests models on its own benchmark and makes new OpenAI models available on launch day, though the report does not provide full benchmark methodology or independently verified results.
What 100 Million Calls Tests
The reported volume illustrates how voice agents are moving from limited trials into routine customer-service operations. At that scale, model performance affects more than the wording of a response: it can shape whether a caller gets an accurate answer, whether an automated system completes a task, and whether a person is brought in when the situation calls for judgment. The reported 80% response share means OpenAI models are part of most calls on Retell’s platform, according to the companies.
Scale also puts pressure on speed and continuity. Retell says daytime traffic can rise to more than 50 times its overnight low. It describes using dedicated servers shared across customers, regional API endpoints and backup systems for telephony, voice and model services. These are company-reported design choices, not evidence that every call meets a particular reliability or latency target.
The customer examples give a more concrete sense of the claimed uses. Electronics company Anker reported an 80.4% case resolution rate for Tier-1 post-sales support in the United States and United Kingdom, along with an NPS of 63. Pine Park Health said its agents recovered medical-assistant time equivalent to 2.1 full-time roles; its scheduling agent booked appointments in 55.7% of cases where it reached a patient. Those results come from the organizations cited in OpenAI’s report and should not be treated as a single, comparable measure across industries.
Products Worth Considering
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
As an affiliate, we earn on qualifying purchases.
How Retell Evaluates Voice Models
Retell says it built a benchmark called vCX to assess models on multi-turn conversations based on production scenarios. The tests cover instruction following, grounding in supplied information, tool selection and natural conversational language. In a benchmark dated October 13, 2026, OpenAI says GPT-6 Astra ranked first for instruction following with a score of 89.7%.
OpenAI also describes earlier evaluations, conducted before GPT-6 Astra launched. In those tests, it says GPT-5.5, with reasoning enabled, scored 88.0% for non-hallucination rate and 88.3% for tool-call correctness, ahead of Gemini-3.1-Pro and Claude-4.6 on those measures. The source says GPT-5.5 also led on non-hallucination rate with reasoning disabled. These are results reported by OpenAI, and the supplied material does not include enough detail to independently compare test design, sample size or statistical uncertainty.
Retell’s use of OpenAI extends beyond deployed voice agents. OpenAI says Retell engineers use Codex to research code, build features and debug issues, and use GPT-6 Astra to test front-end interfaces. The report frames those tools as part of the company’s development workflow, not as a separate measure of call performance.
“We choose models the way our customers choose vendors: on real calls, not demos. OpenAI keeps earning its spot.”
— Bing Wu, Retell AI co-founder and CEO
What the Performance Figures Show
The account does not say whether the 100 million monthly calls figure counts all calls placed through Retell, only completed calls, or another measure. It also does not specify how the reported resolution rate is defined, which customers or period it covers, or how it is calculated. OpenAI attributes the numbers to Retell, but the source material provides no independent audit.
The benchmark scores likewise need to be read as reported results. Details such as the full test set, scoring procedures and margin of error are not supplied here. It is also unclear how the results translate across different customer workflows, languages, call types and production conditions. The report does not provide a breakdown of calls answered by people, calls transferred to staff, or outcomes for customers who did not reach a resolution.
Retell describes redundancy, regional routing and dedicated infrastructure, but the report gives no uptime figures or service-level results. Those details matter when businesses rely on automated agents for healthcare scheduling, financial services or other time-sensitive tasks.
Further Benchmarks and Deployment Data
Retell says it reruns its benchmark as models improve and makes new OpenAI models available when they launch. That could produce additional comparisons, but the value of future results will depend on clear reporting of test methods and whether scores predict outcomes in live customer calls.
The next useful evidence would include a defined call-volume measure, resolution-rate methodology and performance data across customer types, alongside reliability and transfer-to-human figures. Neither the OpenAI report nor the supplied material announces a separate independent review or a specific date for such data. For now, the figures describe the scale and performance Retell and OpenAI report, while broader verification remains outstanding.
Key Questions
How many calls does Retell say its platform handles?
Retell’s monthly volume exceeded 100 million calls by September 2026, according to an OpenAI report. The report says volume was 30 million in early 2026 but does not define precisely how calls are counted.
What share of Retell calls use OpenAI models?
OpenAI says its models generate responses for 80% of calls on Retell’s platform. This is the share reported by the companies, not an independently audited figure in the supplied material.
What does Retell’s reported resolution rate mean?
OpenAI says Retell customers achieve resolution rates above 80%, but the report does not give a definition, time period or customer-by-customer breakdown for that overall figure. The cited customer examples report their own separate outcomes.
Which OpenAI models does Retell use?
The report lists GPT-6.1 Sol, GPT-6 Astra and GPT-5.5 for generating responses, and GPT-5.6 Terra for analyzing transcripts after calls. It also describes access to OpenAI Realtime models and text-to-speech.
Are the call-volume and benchmark figures independently verified?
The supplied source is OpenAI’s account of its work with Retell and attributes the figures to the companies. It does not provide an independent audit or enough benchmark detail to verify the results externally.
Source: openai.com
