Stacks/Customer support

Retell's contact-center voice agents

Retell runs voice agents for more than 4,000 businesses: its own harness handles audio and turn-taking while OpenAI models (GPT-6.1 Sol, GPT-6 Astra, GPT-5.5) generate replies and tool calls, and GPT-5.6 Terra analyzes transcripts afterward. Models are picked with its vCX benchmark of real multi-turn calls. Volume grew from 30M calls a month in early 2026 to more than 100M by September. This is an OpenAI customer story.

Original writeup

openai.com

OpenAI · openai.com

Retell uses OpenAI models to help power 100M monthly calls | OpenAI

GPT‑6 Astra tops Retell’s voice agent benchmark, while Codex helps it reach a $1.6M annual revenue run rate per employee.

Read the writeup

OpenAI (customer story on Retell AI)Published Oct 27, 2026Last verified Nov 13, 2026
Retell harness
GPT-6.1 SolGPT-6 AstraGPT-5.5GPT-5.6 Terra (post-call transcript analysis)OpenAI Realtime models and OpenAI TTS (customer options)GPT-6 Astra computer use (front-end PR testing)
Use case
Voice AI agents for contact centers (questions, actions, and warm handoff to a person) for more than 4,000 businesses, handling 100M+ calls a month
Category
Customer support
Team size
not stated
Orchestrator / framework
During a call, Retell's harness runs the audio models, tracks the conversation, and manages turn-taking and interruptions. OpenAI models use the customer's instructions and context to generate responses and request tool calls, which Retell executes against the customer's systems and can write back to a record. For handoffs the agent briefs the human rep first; after the call GPT-5.6 Terra extracts outcomes.
Memory
not stated
Evals
Retell's vCX benchmark compares dozens of models on difficult multi-turn conversations from real production scenarios (instruction following, grounding, tool calls, natural language) and is rerun as models launch; GPT-6 Astra ranked first for instruction following at 89.7% on the Oct 13, 2026 run.
Observability
not stated
Cost (stated in writeup)
Not stated
Cost per output (stated)
Not stated in the writeup
WhatWeRun estimate

WhatWeRun estimate

Not enough info to estimate

Volume is stated (100M+ calls a month), but customers pick among several models and there's no tokens-per-call figure (the 1 trillion tokens is a cumulative total).

Reviewed Nov 13, 2026

Models
GPT-6.1 SolGPT-6 AstraGPT-5.5GPT-5.6 Terra (post-call transcript analysis)OpenAI Realtime models and OpenAI TTS (customer options)GPT-6 Astra computer use (front-end PR testing)
Observability tools
not stated
Other tools
Dedicated OpenAI servers shared across customersOpenAI regional API endpointsOwn telephony stack with a standby second stackCodex (internal engineering)
Source
https://openai.com/index/retell-ai

OpenAI (customer story on Retell AI) · Published Oct 27, 2026 · Last verified Nov 13, 2026

Related

More like this

decagon.ai

Customer support

Nov 20, 2025Older setup

Decagon's fine-tuned specialist agent models

Customer-facing conversational and voice agents, where Decagon uses a network of specialized models rather than one monolithic model, including for real-time voice where latency matters

Custom multi-model orchestrationAgent Operating Procedures+3
fin.ai

Customer support

Aug 8, 2025Older setup

Intercom Fin

A customer-support AI agent (chat, and in later work email and voice) that Intercom says resolves up to 70% of customers' conversations, with a 99.8% monthly uptime SLA

Custom LLM routing layerBespoke RAG+9Est. $0.09–$0.52 per conversation