OpenAI says Retell now handles more than 100 million calls a month
OpenAI says Retell topped 100 million monthly calls by September, but its usage, resolution, benchmark and revenue figures have not been independently verified.
OpenAI said Retell AI’s voice-agent platform handled more than 100 million calls a month by September, up from 30 million early in 2026. OpenAI said Retell serves more than 4,000 businesses and uses its models to generate responses for 80% of calls.
Those figures provide a company-reported measure of Retell’s production use in contact centers, but they are not independently audited. The available sources do not define what counts as a call or distinguish among attempts, connected calls and completed conversations. Nor do they provide customer-level volume records or an independent check of the 80% model-use share.
Retell’s system handles audio, conversational turn-taking and interruptions. OpenAI language models use customer instructions and conversation context to produce replies or request actions, which Retell carries out through integrations with customer systems. OpenAI said Retell also uses one of its models to analyze transcripts after calls. The company said Retell has consumed more than 1 trillion tokens across OpenAI models, without specifying a measurement period.
OpenAI also said Retell reports customer resolution rates above 80%. A Retell-hosted case study of Anker’s support operation lists an 80.4% case-resolution rate and a customer net promoter score of 63 for Tier-1 post-sales support in the United States and United Kingdom. Those results are vendor-hosted, and the sources do not disclose an independently audited, cross-customer methodology or denominator for Retell’s aggregate rate.
The customer story cited Retell’s October 13 vCX benchmark, in which GPT-6 Astra scored 89.7% for instruction following. Retell’s benchmark methodology says the dataset came from proprietary production call traffic and was graded by a council of language models. Retell says the test focuses on difficult turns and two failure modes, carries judge-model bias and does not measure the full call experience. Those limits complicate comparisons with independently operated evaluations.
OpenAI further said Retell reports an annual revenue run rate of $1.6 million per employee. The sources provide no audited revenue, headcount, calculation date or methodology for that figure.
More news

OpenAI says SentinelOne uses Daybreak for malware analysis

OpenAI’s annualized revenue run rate reportedly nears $70 billion

OpenAI releases cross-country data on ChatGPT work use
