Telnyx

What it is

Telnyx is a communications platform that runs its own private network and holds carrier licences in 30+ countries, so it sells phone numbers, SIP trunking, voice and messaging APIs from its own infrastructure instead of reselling someone else’s. For a support team it shows up in two ways. It is the cheaper carrier under voice-AI platforms: Vapi, Retell AI and Synthflow all document Telnyx numbers or SIP trunks. It also runs voice agents itself, through AI Assistants and, since 6 April 2026, LiveKit on Telnyx. The closest comparison is Twilio, on a lower price list and with a smaller product surface.

Telnyx is private and founder-led, with co-founder David Casem as CEO. We found no acquisition, shutdown or layoff news in the last 12 months.

What changed in 2026

• LiveKit on Telnyx (beta, 6 April 2026). You point an existing LiveKit agent at Telnyx, deploy it as a Docker image, and route a Telnyx number to it. Telnyx runs it on GPUs in 18 points of presence next to the carrier network. Telnyx claims speech-to-text and text-to-speech cost 50% less than on LiveKit Cloud for the same models. That is a vendor claim with no published per-unit rates, and session fees are waived only while the product is in beta.
• Telnyx Flow retired on 3 July 2026. Telnyx dropped its no-code call-flow builder to put its investment into AI Assistants and Voice AI. Flow users had to rebuild on the Voice, Messaging and Chat Completions APIs or move to another orchestration tool.
• MCP in both directions. An AI Assistant can call external MCP servers as tools over HTTP or SSE. Separately, the Telnyx MCP server, a beta since April 2025, exposes numbers, call control, SMS and assistants to Claude or another client.

Pricing reality

The carrier layer is pay-as-you-go with no platform fee. A US number costs $1.00 a month, local or toll-free. Inbound calls on the Voice API cost about $0.0052 a minute, which is $0.0032 for SIP plus a $0.002 Voice API fee. Outbound calls cost about $0.007 a minute and toll-free inbound about $0.017. Twilio charges $0.0085 for inbound, $0.014 for outbound and $2.15 a month for a toll-free number. On the pipes alone, Telnyx is about 40% cheaper for inbound calls and 50% cheaper for outbound. Porting US and Canadian numbers is free.

The Voice AI engine costs $0.05 a minute and covers orchestration, speech-to-text and text-to-speech. Two things are billed on top: telephony, and the LLM by the token. Telnyx estimates an open model on its own GPUs at about $0.006 a minute, while Anthropic and OpenAI models bill at the providers’ own rates. Telnyx puts the all-in cost at about $0.06 a minute. At 100,000 inbound minutes a month that is roughly $5,500 before the LLM. Twilio ConversationRelay comes to about $7,850 for the same minutes. AI minutes bill in 60-second increments, so a 3-minute-20-second call is charged as 4 minutes. If your calls are short, add 10–20% to any per-minute estimate.

Volume discounts require a commitment. Growth takes 15% off telephony for a $2,000 monthly minimum, and Enterprise starts at a $5,000 minimum. The Growth discount applies to telephony only, so at 100,000 inbound minutes it saves about $78 a month. For AI-heavy traffic, the per-minute engine fee sets the bill, not the plan.

Best for

Telnyx fits CX-ops and support-engineering teams that run inbound voice agents above roughly 20,000 minutes a month. At that volume the carrier and per-minute engine costs matter more than orchestration features. It also fits teams already building on LiveKit who want the model, voice and phone line in one region on one bill. As a plain carrier, it is the right choice when you use Vapi or Retell and want the numbers on your own account at lower rates than Twilio.

Do not pick it if you need a packaged contact center with workforce management, QA and supervisor reporting. Telnyx sells parts, not a suite. It is also a poor fit if your launch date cannot absorb a slow account-verification process.

Versus the alternatives

Twilio has the largest share of the CPaaS market, about 25% by Metrigy’s 2025 count, and the broader product line: Flex, Conversation Memory and Intelligence, and a long list of partners. Pick Twilio when you need those layers or one vendor for voice, SMS, WhatsApp and email at enterprise scale. Pick Telnyx when the phone bill is the cost line you are trying to cut.

Infobip and Sinch are the next-largest CPaaS vendors and are strongest in messaging and international SMS. Pick one of them when the project is mainly about messaging rather than voice agents. Among the vendors Metrigy tracked, Bandwidth grew fastest in 2025. It is the alternative US carrier-owned network to consider if Telnyx’s verification or support process stalls you.

Vapi and Retell AI sit above the carrier. Pick them over Telnyx AI Assistants when you want mature call testing, analytics and easy switching between components. You can still put a Telnyx number underneath them.

If none of these fits, run your highest-volume intent for a month on Telnyx AI Assistants, then run the same intent on Vapi over a Telnyx trunk. Compare cost per resolved call, not cost per minute.

Watch-outs

• Account verification can block launch. Reviewers on G2 and Trustpilot report slow or denied verification and suspensions. In February 2025 the FCC proposed a $4.49M penalty over Telnyx’s know-your-customer controls, which Telnyx disputes. Guard: open and verify the account, and port a test number, at least three weeks before go-live. Keep your existing carrier live until the first production week is clean.
• Outages hit Voice AI as well. Third-party trackers logged an API gateway incident of about 12 hours on 23 February 2026, an API outage in one data centre on 21 May, and a partial Voice AI and portal outage on 15 July. Guard: configure a failover route that sends calls to a human queue or a backup SIP trunk, and alert on rising call-setup failures rather than waiting for the status page.
• HIPAA scope is not explicit for agents. Telnyx offers a BAA covering voice, messaging, fax and “AI inference”, but its compliance page does not name Voice AI Assistants. Guard: get Voice AI Assistants named in writing in the BAA before any agent handles PHI.
• Products get cut. Flow went from live to retired, and LiveKit on Telnyx is still a beta with a vendor-only cost claim. Guard: keep agent logic in your own service or in portable LiveKit code, so moving off a Telnyx surface means changing a trunk, not rewriting the agent.