By Sagar Shankaran, Founder of CallSphere
Coturn versus Cloudflare TURN versus LiveKit TURN: real numbers, per-minute cost, and the break-even point for self-hosting AI voice TURN in 2026.
Key takeaways
The TURN question for AI voice is not which is better — it is where the break-even point sits in 2026. Below ~5M minutes per month, managed TURN wins. Above that, Coturn on dedicated boxes wins on cost. The shape changes by region and codec.
A TURN server relays both directions of media for a session. For a single Opus voice agent call that is roughly 60–100 kbps each way, or ~45 MB of egress per 10-minute relayed call. Multiply by your relay rate (15–25% — see the NAT post) and TURN egress is usually the largest single bandwidth line item in a voice deployment.
Three credible options in 2026:
A fourth pattern is an emerging one: dual-track, where you run Coturn for predictable cost and use a managed service as overflow for the long tail. Several large operators run this in 2026 to hedge both directions.
```mermaid flowchart LR User -->|relay| Anycast[Cloudflare anycast TURN] User2 -->|relay| Coturn[Coturn fleet - 3 regions] Anycast --> Agent Coturn --> Agent ```
The crux is the long-tail user: a Tokyo user relayed via a US-east Coturn pays the trans-Pacific RTT twice. Anycast TURN puts the relay 13 ms from the user, every time.
Assume 1M minutes/month of relayed Opus voice, 60 kbps each way = 90 GB/mo egress.
Hear it before you finish reading
Talk to a live CallSphere AI voice agent in your browser — 60 seconds, no signup.
Below ~5M minutes Cloudflare wins outright. Between 5–50M minutes the answer depends on whether your team has the expertise to operate Coturn at three regions reliably. Above 50M minutes Coturn on bare metal wins. Above 500M minutes you should be talking to multiple bare-metal providers and probably running TURN on the same boxes as your SFU to amortize.
CallSphere runs Cloudflare Realtime TURN as the primary across the six verticals (real estate, healthcare, behavioral health, legal, salon, insurance). Reasons:
We keep a Coturn fallback in two regions for the cases where customers require dedicated TURN under their own BAA. Across 37 agents, 90+ tools, 115+ database tables this dual-track setup keeps SOC 2 + HIPAA reviewers happy. Pricing tiers (Starter $149, Pro $499, Enterprise $1499, 14-day trial) bundle the right TURN for each segment. Affiliates earn 22% — see /affiliate.
```ts // Next.js API route, server-only import crypto from "crypto";
export async function GET() { const ttl = 300; // 5 min const username = `${Math.floor(Date.now() / 1000) + ttl}:cs`; const credential = crypto .createHmac("sha1", process.env.TURN_SECRET!) .update(username) .digest("base64");
return Response.json({ iceServers: [ { urls: "stun:stun.cloudflare.com:3478" }, { urls: ["turn:turn.cloudflare.com:3478", "turns:turn.cloudflare.com:5349"], username, credential, }, ], }); } ```
Does Cloudflare see my media? Yes — TURN by definition relays plaintext SRTP packets. Use SFrame for E2EE.
Is Coturn still viable? Yes for high-volume operators willing to run boxes.
Still reading? Stop comparing — try CallSphere live.
CallSphere ships complete AI voice agents per industry — 14 tools for healthcare, 10 agents for real estate, 4 specialists for salons. See how it actually handles a call before you book a demo.
What about Twilio TURN? Functional but priced for the Twilio bundle; rarely competitive standalone.
Does WebRTC over QUIC remove TURN? Not yet — MoQ is still draft, see the MoQ post.
Can I run TURN on the SFU host? Yes — co-locating Coturn and an SFU on the same box saves egress for relayed flows. Common at scale.
What is the bandwidth cost on cellular? A 10-minute relayed Opus call costs ~45 MB; multiply by the relay rate to size your fleet.
Can I bring my own anycast? Possible but extremely operationally heavy. Few teams do it below 100M minutes/mo.
Does TURN cost vary by region? Yes — egress out of Asia-Pacific cloud regions can be 4x US-east. Plan accordingly.
Three rules for TURN sizing in production:
We also keep a 30-day TURN-relay-rate report broken down by ISP and country. The flat lines are boring; the spikes are the early-warning system for carrier issues.
See it live on /demo, pricing in /pricing, or start a /trial.
Written by
Sagar Shankaran· Founder, CallSphere
Sagar Shankaran is the founder of CallSphere, where he builds production AI voice and chat agents deployed across healthcare, hospitality, real estate, and home services. He writes about agentic AI, LLM engineering, and shipping voice agents that handle real calls in production.
See how AI voice agents work for your industry. Live demo available -- no signup required.
BrowserStack offers 30,000+ real devices; Sauce Labs ships deep Appium automation. Here is how AI voice agent teams use both for WebRTC mobile QA in 2026.
WebTransport is Baseline as of March 2026. Media Over QUIC ships in production within the year. Here is what changes for AI voice agents — and what stays the same.
On May 4 2026 OpenAI published its Realtime stack rebuild — split-relay plus transceiver edge. Here is what changed and what it means for production voice agents.
Evaluate build vs buy for enterprise calling platforms. Architecture patterns, SIP infrastructure, WebRTC, cost models, and timeline estimates for custom telephony systems.
Each Cloudflare agent runs on a Durable Object with its own SQLite, WebSockets, and scheduling. Agents Week 2026 shipped MCP, Code Mode, and 10GB SQLite per agent.
Live news studios in 2026 deploy an AI fact-checker behind every anchor, validating claims against trusted sources and offering on-air corrections within 30 seconds. Here is the production stack.
© 2026 CallSphere LLC. All rights reserved.
Watch how CallSphere handles real customer calls, schedules appointments, and processes payments — live.