Short answer: On 29 October 2026, voice-model company Cartesia launched Managed Agents, a hosted platform for enterprise phone and voice agents. It costs $0.03 per minute, plus $0.014 per minute on a Cartesia phone number, and LLM usage is billed separately at token prices (the launch-time free-LLM offer is no longer listed). That works out to about $44 per 1,000 minutes before LLM costs. Cartesia says it ranks #1 of 30 agents on Artificial Analysis’s Speech Agent Arena.
Prices come from Cartesia’s pricing page and docs, checked 31 October 2026 and re-checked 11 November 2026 (LLM billing and concurrency updated). Benchmark results are as published by Cartesia. We haven’t tested the platform.
Cartesia is best known for its Sonic text-to-speech and Ink speech-to-text models, which other voice agent platforms also offer (Retell, for example, lists Cartesia voices as an option). With Managed Agents, it now sells the whole agent: listening, turn-taking, reasoning, tool calls and speech, run as one hosted system. For B2B teams, that adds a fourth serious option to the platforms we compared yesterday in Vapi vs Retell AI vs Bland AI, and it comes in with a much lower per-minute platform fee.
What Cartesia launched
According to Cartesia’s launch post, Managed Agents is an “end-to-end voice agent platform for enterprise conversations”. Cartesia says it designed the agent harness and its voice models together, so turn-taking, interruptions and tool calls are tuned as a single system rather than a chain of separate vendors. It also says the platform has “already powered tens of millions of conversations in finance, retail, hospitality, and healthcare”.
The launch includes:
- Multilingual agents: 44 languages, with mid-call switching between English, Spanish, Hindi, French and Japanese today, and “all 44 coming soon”.
- Tool calls: agents can call your APIs, CRM and systems of record during a call.
- Knowledge bases: answers from “hundreds of documents”, with context prefilled so the agent doesn’t pause to search.
- Outbound campaigns: batch calling with per-recipient personalisation. The batch calling docs allow up to 5,000 recipients per batch.
- Managed LLMs: models from OpenAI, Anthropic and Google Gemini, with no provider account or API key needed, and automatic fallback to another model if one fails (docs).
- Evaluations: built-in and custom metrics on completed calls, plus agent versioning with restore (docs).
How much does Cartesia Managed Agents cost?
Managed Agents is billed per minute in US dollars, separately from the credits that Cartesia’s speech APIs use. From the pricing page and the pricing docs:
| Item | Price |
|---|---|
| Agent calling (base rate) | $0.03 per minute |
| Telephony on a Cartesia-provided number | +$0.014 per minute |
| LLM usage during calls | Billed separately at the model’s token prices (free-LLM launch offer no longer listed as of 11 Nov 2026) |
| US phone numbers | No charge for the number itself; 1 slot (Free), 3 (Pro), 5 (Startup), 10 (Scale) |
| Concurrent calls | 8 (Free), 12 (Pro), 20 (Startup), 60 (Scale), custom on Enterprise |
Each subscription includes the same amount in prepaid agent dollars. Cartesia’s FAQ converts that into minutes at $0.03 per minute, before telephony and LLM costs:
| Plan | Price per month | Prepaid agent usage | About this many agent minutes |
|---|---|---|---|
| Free | $0 | $1 | 33 minutes |
| Pro | $5 | $5 | 2 h 47 min |
| Startup | $49 | $49 | 27 h 13 min |
| Scale | $299 | $299 | 166 h 7 min |
| Enterprise | Custom | Custom | Custom concurrency, DPAs and BAAs, SSO |
If you use up the prepaid dollars, calls keep running at your plan’s voice agent rate. Two limits matter for non-US teams: Cartesia numbers are US numbers only and can only dial US destinations. To call other countries, import numbers from Twilio or connect a SIP trunk, which your own carrier bills.
How does it compare with Vapi, Retell and Bland?
Here’s 1,000 minutes a month, using the rates we checked on 30 October for our voice AI comparison and Cartesia’s rates from today.
| Platform | ~Cost per 1,000 min | Includes LLM? | Includes telephony? |
|---|---|---|---|
| Cartesia Managed Agents | $44 | No (LLM billed separately) | Yes, on a Cartesia US number |
| Vapi | $82–$129 | Yes (at cost) | Vapi numbers free; carrier extra if you bring one |
| Retell AI | ~$127 | Yes (calculator default) | Yes, Retell US Twilio at $0.015/min |
| Bland AI (Start) | $140 + telephony | Yes | No |
Our arithmetic from official list prices. This isn’t a like-for-like comparison: Cartesia’s rate excludes the LLM, while the other three already include one. A small model adds little per minute; a frontier model can add more than the whole platform fee.
The takeaway for buyers: Cartesia is currently the cheapest way to run a fully hosted voice agent on list prices, and it stays competitive if you choose a small, fast model. The main question is how much the LLM will add, and you can check that through the models endpoint, which returns token prices for each model available to your account.
What do the benchmarks say?
Cartesia’s launch post cites two benchmarks, both run with GPT-5.5 as the reasoning model:
- Speech Agent Arena (Artificial Analysis): blind human raters compare agents on tasks such as booking appointments. Managed Agents ranks #1 of 30 with a preference Elo of 1115, ahead of Gemini 3.1 Flash Live Minimal (1099) and Gemini 3.8 Live (1079).
- EVA-Bench (ServiceNow’s open benchmark): first-attempt scores across 200+ customer service, IT service and HR scenarios. Managed Agents scores 87.4% accuracy and 90.7% experience, against 73.7% and 71.2% for the next system on accuracy, xAI’s Grok Voice 3.0.
Cartesia 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. That figure has no published method, so treat it as a vendor claim.
Why this matters for B2B automation
- Voice-agent platform fees are falling. A model maker selling the full stack at $0.03 a minute puts pressure on platforms that resell its own speech models. That could pressure the platforms that resell these models on top of their own fees.
- The LLM becomes the main cost. When the platform fee drops to a few cents, model choice drives the bill. Short prompts, small models for routine calls and a bigger model only for complex cases will matter more than which platform you pick.
- Bundled harnesses are winning. Like Microsoft’s Copilot harness in Foundry, which we covered in this week’s AI tools roundup, Cartesia now runs the agent loop for you. You trade some control for faster setup and fewer vendors to manage.
How to connect Cartesia agents to n8n
Cartesia gives agents three kinds of tools: system tools (end the call, send DTMF tones, transfer to a phone number), webhook tools that call your server, and client tools that run in your own app over WebSocket. For CRM lookups and bookings, webhook tools are the simplest route to n8n:
- Create a webhook tool with your n8n Webhook node production URL as
api_schema.url. It must be HTTPS, and Cartesia blocks private and internal network ranges, so a LAN-only n8n won’t work. - Use
GET,POST,PUT,PATCHorDELETE, and add bearer authentication. Check that token in n8n before the workflow does anything. - Return the answer with a Respond to Webhook node. Cartesia waits 20 seconds by default (configurable from 1 to 120), so keep live-call lookups fast and push slow jobs to a separate workflow.
- Return only the fields the agent needs; short responses keep the call responsive.
- For post-call automation, register a webhook in Cartesia’s API to receive call events for each agent, then update your CRM, send a follow-up or alert a human.
Before you expose workflows to a voice agent, read our guide to securing your n8n instance and run the workflow through the free n8n Workflow Auditor. Estimate execution costs for high call volumes in the n8n Cost & Hosting Planner, and compare the AI subscriptions in your stack with the AI SaaS Pricing Comparison.
What we couldn’t verify
- How much LLM usage adds per minute for the default model. Token prices are returned by the models API for logged-in accounts and aren’t published. (The launch-time free-LLM offer was no longer listed on 11 November 2026.)
- How concurrency is enforced in practice. Since our first check, the pricing page publishes concurrent-call limits (8 / 12 / 20 / 60 from Free to Scale); we haven’t load-tested them.
- Benchmark independence. Rankings come from third-party benchmarks but are reported by Cartesia. We haven’t run test calls.
- The “tens of millions of conversations” claim, which comes from Cartesia’s launch post.
Frequently asked questions
What did Cartesia launch on 29 October 2026?
Cartesia launched Managed Agents, a hosted platform for enterprise voice agents. It bundles Cartesia’s speech models with a managed agent harness, telephony, knowledge bases, tool calls, outbound batch calling and call evaluations, so teams don’t have to stitch together separate speech-to-text, LLM and text-to-speech providers.
How much does Cartesia Managed Agents cost?
Cartesia’s pricing page and docs list $0.03 per minute for agent calls, plus $0.014 per minute when you use a Cartesia-provided phone number. LLM usage is billed separately at the chosen model’s token prices. (At launch the pricing page said LLM usage was free for a limited time; on 11 November 2026 it says “Billed separately”.)
Is there a free plan?
Yes. The Free plan includes $1 of prepaid agent usage a month, which Cartesia says covers about 33 minutes at $0.03 per minute, plus one US phone number slot. Paid plans are Pro ($5 a month), Startup ($49) and Scale ($299), and each includes the same dollar amount as prepaid agent usage.
Is Cartesia cheaper than Vapi, Retell or Bland?
On the platform fee alone, yes: $0.044 per minute with Cartesia telephony, against roughly $0.08 to $0.14 per minute on the other three. But Vapi, Retell and Bland figures include an LLM, and Cartesia’s long-term rate excludes it. Add your model’s token cost before you compare.
Can Cartesia agents call my n8n workflows?
Yes. Webhook tools send an HTTPS request to your endpoint during a call and return the response to the model, with a default 20-second timeout (configurable from 1 to 120 seconds). Point the tool at an n8n Webhook node and reply with a Respond to Webhook node. Cartesia blocks private and internal network addresses, so the endpoint must be public HTTPS.
Are the benchmark results independent?
Partly. The rankings come from third-party benchmarks (Artificial Analysis’s Speech Agent Arena and ServiceNow’s open EVA-Bench), but Cartesia published the results in its own launch post, run with GPT-5.5 as the reasoning model. The 85% vs 67% turn-taking figure is Cartesia’s internal evaluation. Test with your own calls before you switch.
Related reading
- ElevenLabs vs Cartesia: TTS and voice agent pricing compared
- Vapi vs ElevenLabs Agents: voice AI pricing and concurrency
- Vapi vs Retell AI vs Bland AI: voice agent pricing compared
- WhatsApp Business API pricing changes, October 2026
- Lindy vs Relevance AI vs n8n for B2B AI agents
- ServiceNow and OpenAI: ServiceNow workflows from ChatGPT
Sources
Cartesia: Introducing Managed Agents (29 October 2026); Cartesia pricing and pricing docs; Cartesia docs on Managed Agents, Managed LLMs, tools, webhook tools, Cartesia numbers and batch calling; n8n docs for the Webhook and Respond to Webhook nodes. Vapi, Retell and Bland costs come from their official pricing pages, checked 30 October 2026. All pages checked 31 October 2026 unless noted.
