Cartesia Launches Managed Agents, an End to End Voice Agent Platform for Enterprises
Introducing Cartesia Managed Agents: our end-to-end voice agent platform for enterprise. #1 on @ArtificialAnlys Speech Agent Arena and @ServiceNow EVA-Bench, built by co-designing our voice models and harness as one system. Check it out →
Context
Cartesia says designing its voice models and agent software together puts it first on two independent voice agent benchmarks, a sign that tight integration may beat stitched together pipelines.
Cartesia is a San Francisco company known for fast text to speech models. Voice agents answer and make phone calls for businesses. Artificial Analysis runs a Speech Agent Arena where people compare agents blind, and ServiceNow publishes EVA Bench, an open benchmark for enterprise voice agents.
Deep dive
Cartesia said on Thursday it is introducing Managed Agents, an end to end voice agent platform for enterprise. According to the post, it ranks first on the Artificial Analysis Speech Agent Arena and ServiceNow's EVA Bench, which Cartesia credits to building its voice models and the agent harness as one system.
Cartesia's blog says listening, turn taking, reasoning, tool calls and speech are tuned together, and that the platform includes telephony, knowledge bases and tool integrations. It says the product has already handled tens of millions of conversations in finance, retail, hospitality and healthcare.
The blog reports a preference Elo of 1115 on the Speech Agent Arena, first of 30 agents, ahead of Gemini 3.1 Flash Live Minimal at 1099 and Gemini 3.8 Live at 1079. On EVA Bench it reports 87.4 percent accuracy and 90.7 percent experience scores, ahead of Grok Voice 3.0 at 73.7 and 71.2. Cartesia notes the benchmarks were run with GPT 5.5 as the reasoning model, so part of the result depends on OpenAI's model.
The company also says agents support 44 languages, with switching between five of them mid call for now. The internal turn taking figures it cites are its own.
What to expect next
Watch whether the rankings hold as Google, OpenAI and xAI update their live voice models, and as Phonic, which launched Phonic 1 the same day, competes for the same customers. Enterprise buyers will look at reliability on long, complex calls and pricing, which the blog did not list. Support for all 44 languages mid call is promised but not yet available.