Cartesia Launches Managed Agents, an End-to-End Voice Agent Platform

One system for listening, reasoning and speech

Cartesia says it designed the agent harness and its voice models together, so listening, turn-taking, reasoning, tool-calling and speech are tuned as one end-to-end system. The platform also bundles what enterprise teams need to run agents in production, including telephony, knowledge bases and tool integrations.

The company argues that existing approaches each have a weakness. Cascaded pipelines are the dominant paradigm for enterprise tasks, but without careful co-design they can lose out on naturalness and conversational dynamics such as turn-taking and interruptions. Today's speech-to-speech models sound natural, Cartesia says, but they are hard to control and less reliable for enterprise-grade tasks. Managed Agents is meant to deliver a natural conversation without giving up the intelligence and controllability complex tasks require.

According to Cartesia, the platform has been tested at scale and has handled tens of millions of conversations in finance, retail, hospitality and healthcare.

Benchmark results

On Artificial Analysis's Speech Agent Arena, blind human raters compare agents on tasks such as booking appointments and ordering takeout. Managed Agents ranks #1 of 30 agents with a preference Elo of 1115. In the October 2026 listing, Google's Gemini 3.1 Flash Live Minimal follows with 1099 and Gemini 3.8 Live with 1079. Grok Voice 3.0 from xAI scores 1064, OpenAI's GPT-Live-1 variants score 1041 and 1037, and ElevenLabs entries score 1003 and 989.

On EVA-Bench, ServiceNow's open benchmark for enterprise voice agents in CSM, ITSM and HR, Cartesia reports the highest scores for both accuracy (87%) and experience (91%) across 200+ scenarios, which it describes as a new Pareto frontier. The detailed table lists 87.4% accuracy and 90.7% experience for Managed Agents. Grok Voice 3.0 posts 73.7% and 71.2%, and Gemini 3.8 Live posts 45.9% and 88.6%. Cartesia notes that its benchmarks were run with GPT-5.5 as the reasoning model.

Nitin Aggarwal, Senior Director of Product Management, Agentic Evals and RAI, is quoted in the announcement saying Managed Agents "posted one of the strongest results we've seen to date" on EVA-Bench.

In Cartesia's own internal evaluations, callers reported no turn-taking issues in 85% of conversations with Managed Agents, compared with 67% for other agents.

Built for complex calls

Cartesia positions the platform for calls that rarely stay simple, such as claims intake, patient scheduling, retail returns and IT helpdesk. One conversation can involve verifying a caller, checking three systems and quoting policy. The listed features are:

• Multilingual: agents can be built in 44 languages. They can switch between English, Spanish, Hindi, French and Japanese mid-conversation, with all 44 coming soon.
• Tool calls: backend APIs, CRMs and systems of record can be integrated so an agent can look up records and take action mid-call.
• Knowledge bases: agents can give precise answers from hundreds of documents, with context prefilled dynamically so they respond without pausing to search.
• Outbound campaigns: thousands of calls can be scheduled in outbound batches, each personalized to the person being called.

What it means for enterprise teams

For companies deploying voice AI on customer calls, the pitch is a single platform instead of separately assembled speech, reasoning and telephony components. The benchmark figures come from Artificial Analysis and ServiceNow's EVA-Bench, though the turn-taking numbers are from Cartesia's internal tests, and all results were produced with GPT-5.5 as the reasoning model. Cartesia's launch post does not give pricing or availability details.