Cartesia’s new Managed Agents voice platform is listed first on Artificial Analysis’s Speech Agent Arena, a blind human-preference ranking, with an Elo of 1111 when we read the board on 30 October. Google’s Gemini 3.1 Flash Live Minimal is second at 1089, and OpenAI’s best entry, GPT-Live-1 (Sol, low), is sixth at 1037.
The lead is narrow. By our arithmetic, Cartesia is 22 points ahead of Google, and the board’s 95% intervals overlap: 1083 to 1140 for Cartesia against 1064 to 1114 for Gemini. The board gives Cartesia a rank range of 1 to 2. On ServiceNow’s EVA-Bench, Cartesia puts its accuracy at 87.4%; by our arithmetic, that is 13.7 points above the next-highest accuracy in Cartesia’s table.
Business Pill · The Model and the Harness
An evergreen Business Engineer explainer on models and harnesses. It explains the general idea only and does not discuss Cartesia.
The key insight: Cartesia’s Managed Agents is listed first on Artificial Analysis’s blind Speech Agent Arena at 1111 Elo, 22 points ahead of Google’s Gemini 3.1 Flash Live Minimal by our arithmetic, with overlapping 95% intervals.
Cartesia announced Managed Agents on 29 October in a blog post, calling it “our end-to-end voice agent platform for enterprise conversations”. The company says it “designed the harness and our voice models together”, so that listening, turn-taking, reasoning, tool-calling and speech are tuned as one system.
Cartesia says the platform “has already powered tens of millions of conversations in finance, retail, hospitality, and healthcare”. It lists telephony, knowledge bases, tool integrations and outbound call campaigns, and says agents can be built in 44 languages.
Cartesia’s launch post on X, above, says the product is “#1 on @ArtificialAnlys Speech Agent Arena and @ServiceNow EVA-Bench”.
Artificial Analysis Speech Agent Arena, Elo with 95% interval, read 30 Oct 2026: Cartesia Managed Agents 1111 (1083-1140), Gemini 3.1 Flash Live Minimal 1089, Gemini 3.8 Live 1074, Grok Voice 3.0 1059, GPT-Live-1 (Sol, low) 1037, ElevenLabs Agents default cascaded 985. The 22-point gap is our arithmetic. Source: Artificial Analysis.
The Two Rankings
On the Speech Agent Arena, Cartesia’s blog says blind human raters compare agents on tasks “like booking appointments and ordering takeout”, and that Managed Agents “ranks #1 of 30 agents with a preference Elo of 1115”. The live board we read a day later showed 1111, plus or minus 29. Behind three Google Gemini entries, xAI’s Grok Voice 3.0 was fifth at 1059 and ElevenLabs’ default cascaded system was 12th at 985.
On EVA-Bench, which Cartesia describes as ServiceNow’s open benchmark for enterprise voice agents in customer service, IT and HR, the blog says Managed Agents scored 87.4% on accuracy and 90.7% on experience across “200+ enterprise scenarios”. The next-highest accuracy in Cartesia’s table is Grok Voice 3.0 at 73.7%; the next-highest experience score is Gemini 3.8 Live at 88.6%, whose accuracy is listed at 45.9%.
Cartesia quotes Nitin Aggarwal, a ServiceNow product director, saying the result “reflects a broader pattern we’re seeing in enterprise voice: harnesses and models designed together”.
Pipeline Versus Speech-to-Speech
Cartesia frames the launch as a contest of designs. Its blog says “Cascaded pipelines are the dominant paradigm for enterprise tasks” but can lose on naturalness, while “Today’s speech-to-speech models sound natural, but they’re hard to control and less reliable for enterprise-grade tasks”. Rohan Paul, in a post on X, describes Managed Agents as a cascaded voice agent that won against native speech-to-speech models.
Cartesia says its benchmarks were “run with GPT-5.5 as the reasoning model”, so the language model inside the leading agent comes from OpenAI. It also says that in its internal evaluations, callers reported no turn-taking issues in 85% of conversations with Managed Agents, against 67% for other agents.
As we read Cartesia’s pitch, the claim is about design, not a single model: the company says it designed the harness and its voice models together.
As we read the EVA-Bench table Cartesia published, the gap is wider on accuracy than on experience: 13.7 points on accuracy, by our arithmetic, against 2.1 points on experience over Gemini 3.8 Live.
As we read the benchmark note, the reasoning inside the leading agent is OpenAI’s: Cartesia says the tests were run with GPT-5.5 as the reasoning model.
Nitin Aggarwal, ServiceNow, quoted by Cartesia
“harnesses and models designed together”
Three Implications
FOR ENTERPRISE BUYERS Cartesia says Managed Agents has already powered tens of millions of conversations in finance, retail, hospitality and healthcare.
FOR VOICE MODEL VENDORS On the live board we read, Google’s Gemini 3.1 Flash Live Minimal sits 22 points behind Cartesia, and its 95% interval overlaps Cartesia’s.
FOR MODEL LABS Cartesia says its benchmarks were run with GPT-5.5 as the reasoning model, so OpenAI’s model sits inside the agent listed first.
The framework argues that “the conversational interface was scaffolding. The harness is the building.”
As we read it, Cartesia’s launch makes the same argument for voice: the company says it designed the harness and its voice models together. The framework was published before this launch and does not discuss Cartesia.
What Is Not Established
The Speech Agent Arena ranking is a live Elo board and moves as votes arrive; Cartesia’s blog figure of 1115 and the 1111 we read differ for that reason, and the first two places sit within overlapping intervals. The EVA-Bench table we cite is the one Cartesia published, and the 85% turn-taking figure is from Cartesia’s internal evaluations. Cartesia did not publish pricing in the launch post we read. The 22-point and 13.7-point gaps are our arithmetic.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
Cartesia’s Managed Agents is listed first on Artificial Analysis’s blind Speech Agent Arena, 22 Elo points ahead of Google within overlapping intervals. Cartesia’s published EVA-Bench table puts its accuracy 13.7 points above the next system, by our arithmetic, in tests Cartesia says were run with GPT-5.5 as the reasoning model.
A note on sourcing. Launch details, the EVA-Bench table and the turn-taking figure come from Cartesia’s blog post of 29 October 2026. Arena figures are from Artificial Analysis’s live Speech Agent Arena, read on 30 October 2026. The 22-point and 13.7-point gaps are our arithmetic. We did not contact Cartesia.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
