Skip to content

Retell surpasses $100M in annualized run rate as it leads G2’s CX and Voice AI Agent leaderboards

Andrew BlairEvaluations

Updated

Retell’s $100M annualized run rate milestone, with a blue voice waveform representing its growth in enterprise voice AI.

Retell has surpassed $100 million in annualized run rate, less than three years after its public launch in February 2024. The company grew from $10 million to $100 million in 17 months and serves more than 4,000 businesses in production.

The milestone arrives as Retell is currently ranked first on G2’s CX AI agent leaderboard and its Voice AI agent leaderboard (as of November, 2026). The two rankings give buyers a closer look at the agent’s performance: G2’s CX AI agent evaluation tests customer-support workflows, while Bluejay’s Voice AI agent evaluation, published by G2, tests phone-based tasks.

Retell’s monthly call volume more than tripled from 30 million in early 2026 to over 100 million by September, according to an OpenAI customer story. 

Enterprise growth accelerates Retell’s customer base

Retell is a next-generation AI voice agent platform for contact centers, helping enterprises build, deploy, and continuously improve AI agents at scale. Its agents answer questions, use connected business systems, update records and transfer callers to a person, while Retell manages conversation flow, interruptions and tool execution.

Retell supports appointment setting, debt collection, inbound and outbound sales, and cites its top use cases in healthcare, insurance and financial services. According to Retell, at Fortune 500 automotive retailer Asbury Automotive Group, Retell agents book over 60,000 service appointments monthly, helping the company save $2 million annually. For the San Antonio Spurs, Retell agents handle season ticket renewal calls, delivering a 505× return on investment.

In a recent interview with Selina Wang, Retell GTM lead, she described the company was originally built around a self-serve, developer-first model and is now seeing increased enterprise adoption, with enterprise revenue growing ninefold in 2026. Customers include Lenovo, BILL, and CVS Health. As it expands upmarket, Retell continues to emphasize platform openness, configurability, and customer control.

As enterprises move AI agents from pilots into production, the ability to reliably complete real-world tasks becomes increasingly important. Retell’s performance across both evaluations offers independent evidence of its ability to handle complex customer-facing workflows.

“We believe that the people who know your customers best should be the ones shaping the AI that serves them,” says Bing Wu, co-founder and CEO. “Real transformation happens when enterprises can build, deploy, and continuously improve AI agents themselves. We’re giving them the tools and control to make that possible.”

G2’s CX AI agent leaderboard puts Retell first on task success

Retell leads the current CX AI agent leaderboard with 243 out of 250 task-success points, alongside 90% accuracy and 92% policy compliance. Yuma and LiveAgent follow with displayed task-success scores of 240. Yuma records 97% accuracy and 95% compliance, illustrating why buyers should examine the individual measures as well as the ranking.

The CX AI agent evaluation gives each product the same 50 support tasks inside a simulated company, with a written policy and 38 working business tools. Those tools support actions such as refunds, subscription changes and account updates.

The evaluation examines the conversation, the tools the agent used and the state of the business systems afterward. A helpful-sounding response is insufficient if the correct action never occurs.

Task-success points reward completed work. Accuracy and policy compliance are reported separately, helping buyers distinguish whether an agent finished the task from how reliably it followed the facts and rules along the way.

Bluejay’s Voice AI agent evaluation ranks Retell first on composite score

In Bluejay’s Voice AI agent evaluation, published by G2, Retell leads with a composite score of 81.4 out of 100, followed by ElevenLabs Agents at 80.9 and LiveKit Agents at 80.4.

Its underlying results show an 85.6% task-success rate, 4.7% word error rate on clean audio and 2.15-second median response latency. Those measures reveal differences the overall ranking alone cannot: LiveKit records higher task success at 89.1%, while ElevenLabs responds faster at a 1.15-second median and records a lower clean-audio word error rate of 3.7%.

Bluejay evaluated six platforms using 90 simulated tasks, each run five times per platform. Bluejay drew its tasks, simulated business environments and grading criteria from its MIVAS benchmark. The scenarios cover a dermatology front desk, legal intake and electronics customer support, with additional tests under degraded audio conditions.

The same agent instructions and tool schemas are deployed across platforms. A call passes when the required tools are used correctly and the resulting business records match the expected outcome. Repeating tasks also lets buyers examine consistency across runs, rather than relying on one successful conversation. Read the Voice AI agent methodology.

Evaluation results guide buyer shortlists and workflow testing

Retell’s revenue and call-volume growth show the scale at which businesses are using its platform. As agents take on more customer-facing work, buyers need evidence of what they can complete reliably and what happens when they fail.

“Enterprise buyers compare several platforms, independent evaluation results help them build the shortlist faster. The best AI teams then go deep on the tasks they failed: was it a memory issue, context issue, or because of too many tool calls? Do we need to switch to a better model? They can fix issues quickly and hit the bar they need on most workflows. When we evaluated Retell it was one of the quickest platforms for our engineers to set up, where others take up to 50 hours. This means pilots can start in a few days and the team can quickly demonstrate where their product adds value to the business.” – Cyril Delattre, Agents Evals Product at G2

Buyers can use the results to build a shortlist and identify which tradeoffs to test against their own workflows, policies and integrations. Those tests matter because the published results reflect specific configurations and conditions.

Comments

Questions about how this works? Ideas for what comes next? Join the discussion.

Loading comments…