Verdict
Telnyx suits developers who want the agent, phone numbers and calls on one invoice at a low public per-minute price, with tools, MCP servers and warm transfers built in 1,3,6. The catch: outbound calls are scheduled one recipient at a time, training opt-outs go through the account team, and Trustpilot reviewers report slow verification and support 5,9,18.
Good fit if you are
- Teams that want phone numbers, calling and the AI agent from one vendor on one bill
- Developers who want low per-minute costs and open models hosted by the carrier
- Teams taking card payments by phone that must keep card details away from the AI
Look elsewhere if you are
- Teams that need list-based outbound campaigns out of the box
- Buyers who want fast human support from day one
- Companies that need a written no-training commitment without negotiating
Pros and cons
What we like
- Agent, phone numbers and calls run on one network and one invoice 1.
- Low public pricing: a $0.05-a-minute voice engine, with the default LLM and US telephony adding about a cent 1.
- Webhook tools, MCP servers and CRM integrations act in other systems during calls 3.
- Warm transfers let the agent brief a person privately, who accepts before the caller is bridged 6.
- Tests, saved versions and rule-based traffic routing come with the builder 7.
What to watch
- No batch campaign tool is published; scheduled outbound calls go to one recipient at a time 5.
- Guardrails and personal-data redaction for voice calls are not published; only the Pay tool masks card details 10.
- Training opt-outs must be arranged with the account team 9.
- Each call is billed in full minutes, which raises the cost of short calls 1.
- Trustpilot reviewers report week-long verification checks and slow support 18.
Key facts
| Made by | Telnyx |
|---|---|
| Pricing | $0.05/min voice engine + LLM tokens + telephony at carrier cost; no monthly fee on pay as you go |
| Cost for 1,000 minutes | $63 a month (1,000 min × $0.05 voice engine = $50 + default Kimi LLM at Telnyx's estimate of $0.0064/min = $6.40 + US inbound calling 1,000 × $0.0032 = $3.20 + Voice API fee 1,000 × $0.002 = $2.00 + one US local number $1.00 = $62.60; rounding each call up to a full minute can add more) |
| Also requires | A card on file; pay as you go has no monthly minimum |
| Channels | Inbound calls, Outbound calls, Web and app calls, SMS, Chat, WhatsApp |
| Languages | 80+ languages and dialects 2; 3,800+ voices across 80+ language families 8 |
| Works with | Salesforce, ServiceNow, HubSpot, Zendesk, Intercom, Jira, GitHub, Greenhouse, Microsoft Teams, Notion, Stripe |
| Security reports | SOC 2 Type II, ISO 27001:2022, ISO 27701:2019, PCI DSS |
| Free trial | No trial published; free sign-up with pay as you go |
What Telnyx Voice AI Agents is
Telnyx is a communications carrier that sells phone numbers, calling, messaging and AI inference on its own network 2,12. Its Voice AI Agents, called AI Assistants in the developer docs, answer and place phone calls and also handle web calls, web chat, SMS and WhatsApp 3,4. Because Telnyx is the carrier, the agent, the phone number and the call sit on one account and one invoice 1.
Telnyx hosts open models on its own GPUs and offers a no-code builder alongside the API 2,3. In April 2026 it added WhatsApp Business calling, so the same agents can take WhatsApp calls next to phone and browser calls 12.
What it can do
You build an assistant in the no-code AI Assistant Builder or through the API, pick a model and a voice, and attach a Telnyx number 3. New assistants use Kimi K2.6 on Telnyx GPUs unless you choose another model, such as Claude Haiku 4.5, a GPT model or your own endpoint 3,4. Webhook tools call your APIs during the call, MCP servers add tools, and integrations connect Salesforce, ServiceNow, HubSpot, Zendesk and others 3. Transfer and SIP REFER tools pass the call to named targets; with warm transfer acceptance, the agent briefs the person privately and bridges the caller only once they accept 3,6. A Pay tool collects card details by keypad and keeps them out of recordings, transcripts and the AI 10.
Outbound calls start from the API or from scheduled events, which call or text one recipient at a set time and retry unanswered calls 3,5. Before launch, AI Tests run a scenario against success criteria, and you can save versions and route live traffic by rules or percentage splits 7. After calls, conversation history shows transcripts, and AI Insights extract structured data and score instruction-following and satisfaction 3,4.
- Actions. Webhook tools call your APIs during the call, MCP servers add tools, and integrations reach Salesforce, ServiceNow, HubSpot, Zendesk and others 3. Scheduled events place outbound calls or texts at a set time with retries, and a Pay tool takes card payments by keypad 5,10.
- Setup. Build in the no-code AI Assistant Builder in the Mission Control Portal or through the API, attach a Telnyx number, and import existing agents from OpenAI, ElevenLabs, Vapi and other providers 3,4.
- Transfer to people. Transfer and SIP REFER tools send the call to named targets; with warm transfer acceptance, the agent briefs the person privately and bridges the caller only if they accept 3,6.
- Analytics. Conversation history with transcripts, AI Insights for structured post-call analysis with built-in instruction-following and satisfaction scores, and end-to-end tracing with latency breakdowns 2,3,4.
Pricing
Telnyx prices voice AI in layers. The voice engine costs $0.05 a minute and covers orchestration, speech-to-text and text-to-speech 1. LLM tokens are billed per token, and Telnyx estimates its default Kimi model at about $0.006 a minute 1. US calling runs from $0.0032 a minute inbound and $0.005 outbound, plus a $0.002 Voice API fee, and a US local number is $1 a month 1. Each call is rounded up to the next full minute, so short calls cost more 1. Pay as you go has no monthly fee or minimum; Growth, with a $2,000 monthly minimum, cuts voice, messaging and number charges by 15% 1. For 1,000 inbound minutes a month, the total comes to about $62.60 1.
| Plan | Price | Includes |
|---|---|---|
| Pay as you go | $0 platform fee1 | Full platform at standard rates, 500 concurrent calls, 100 API requests per second, community support |
| Growth | $2,000/month minimum spend1 | 15% off voice, messaging and numbers, automatic volume discounts, 24/7 support by chat or call |
| Enterprise | $5,000/month minimum spend1 | Custom pricing, dedicated infrastructure and IP addressing, private network interconnect, no rate limits |
| Voice engine | $0.05 per minute1 | Orchestration, hosted speech-to-text and text-to-speech; each call rounded up to the next full minute |
| LLM tokens | Per token; Kimi typically about $0.006 per minute1 | Open models on Telnyx GPUs at priority-tier rates; Anthropic and OpenAI models at their list rates |
| US telephony | From $0.0032/min inbound and $0.005/min outbound, plus a $0.002/min Voice API fee1 | Calls on Telnyx's own network, on the same invoice |
| US local number | $1.00/month (volume pricing from $0.79)1 | One phone number |
Security and data
Telnyx shows SOC 2 Type II, ISO 27001:2022, ISO 27701:2019 and PCI DSS on its security page 11. HIPAA support is included on every plan, including pay as you go 1, but the BAA terms are not published. Data Locality stores records at rest in the US, EU, APAC or Middle East, but processing for voice calls is not guaranteed to stay in region 9. A per-assistant setting turns off storage of conversation history and insights 9. Training on customer data is left to Telnyx's terms and DPA, and an opt-out must be arranged with the account team 9.
- HIPAA. HIPAA support included on every plan, including pay as you go 1; BAA terms not published
- Data residency. Records stored at rest in the US, EU, APAC or Middle East through Data Locality; processing for voice calls is not guaranteed to stay in region 9
- Training on your data. Governed by Telnyx's terms and DPA; an opt-out from training must be arranged with the account team 9
- AI models. Open models on Telnyx GPUs, with Kimi K2.6 as the default, plus Anthropic, OpenAI or your own OpenAI-compatible endpoint 3,4
What customers say
Telnyx scores 4.7 from 528 G2 reviews and 4.7 from 196 Capterra reviews, but both cover the whole Telnyx platform rather than its voice agents 15,17. G2's separate Voice AI Agents page has only 5 reviews 16. Trustpilot is split at 3.2 from 646 reviews, with complaints about verification delays, account suspensions and slow support 18. Trioteca, a Spanish mortgage broker, runs seven Telnyx voice agents, one of which places about 1,000 outbound calls a day 13. PatientSync's CTO is quoted on the product page 2.
| Platform | Rating | Reviews | Seen |
|---|---|---|---|
| G2 (Telnyx CPaaS (whole platform))15 | 4.7 / 5 | 528 | 8 October 2026 |
| Capterra (Telnyx Suite (whole platform))17 | 4.7 / 5 | 196 | 8 October 2026 |
| Trustpilot (telnyx.com)18 | 3.2 / 5 | 646 | 8 October 2026 |
What Telnyx claims (not independently verified):
- A production agent runs about $0.06 per minute all-in 1
- Sub-200ms latency from GPUs co-located with telephony points of presence 2
- Licensed carrier in 45+ countries 2
- 14,000+ companies build on Telnyx 14
- Trioteca: seven agents running around the clock, one placing about 1,000 outbound calls a day, and revenue doubled year over year 13
Concerns raised by others:
How we scored it
| Criterion | Why | Score |
|---|---|---|
| What it can do 30% of the total | Agents call your APIs and MCP servers on inbound and outbound calls and transfer to a person 3,6, speak 80+ languages 2 and work on SMS and WhatsApp 4,12, but outbound calls are scheduled one recipient at a time and no batch campaign tool is published 5. | 4/5 |
| Pricing 20% of the total | The $0.05 voice engine, the per-minute LLM estimate, US telephony and number prices, concurrency limits and plan minimums are all public, and pay as you go has no minimum 1. | 5/5 |
| Control 20% of the total | Documented: tests before launch and saved versions with traffic routing 7, configurable transfers with warm acceptance 3,6 and conversation history with AI Insights 3,4; guardrails or personal-data redaction for voice calls are not published beyond the Pay tool's card masking 10. | 4/5 |
| Security and data 15% of the total | SOC 2 Type II and ISO 27001 11 plus Data Locality in four regions and HIPAA on every plan 1,9, but training on customer data is left to the terms and DPA 9. | 4/5 |
| Customer evidence 15% of the total | Telnyx scores 4.7 from 528 G2 reviews and 4.7 from 196 Capterra reviews, but both pages cover the whole Telnyx platform rather than its voice agents 15,17; Trioteca runs seven Telnyx voice agents 13. | 4/5 |
| Total | Weighted sum of the 5 levels, out of 10. | 8.4 |
Sources
- AI Voice Agent Pricing
- Voice AI Agents
- Voice Assistant (no-code guide)
- Telnyx AI: Assistants documentation index
- Scheduled Events
- Warm Transfer Acceptance
- Testing, Versions & Traffic Distribution
- Conversational AI quickstart: Models & Supported Languages
- Data Residency & Compliance FAQ
- Pay over Voice
- Security
- Telnyx Launches WhatsApp Business Calling, Bringing Programmable Voice to the World's Largest Messaging Platform
- Why Trioteca Chose Telnyx: Trust
- Customer stories
- Telnyx CPaaS Reviews
- Telnyx Products
- Telnyx Suite Reviews
- Telnyx Reviews