Voice Network Costs and SLAs: What Carriers Sign

Quick answer: Of six carriers whose own pages we could read on 18 October 2026, only Twilio and Sinch publish a voice uptime SLA: 99.95% a month. Twilio’s API SLA pays a single 10% credit, claimed within 30 days of month-end; Sinch promises “reasonable endeavours” and no credit. Neither covers call quality or outside carrier networks. Outbound rates to a US, UK, Australia, India and Germany basket ran from US$0.005 to US$1.07 a minute, depending on where the call starts.

Where Zian sits on top of the carrier network

Zian AI is the conversation layer that runs on top of a voice network, not the network itself, so a carrier’s SLA is a separate line you check whichever agent platform you choose. Zian’s AI receptionist answers inbound calls 24/7 (qualifying, booking, transferring and taking messages), its outbound agents follow up by voice, SMS, email, WhatsApp, Instagram DMs and Facebook Messenger, and it handles answering machines and voicemail on calls, in 30+ languages.

zian.ai does not publish which carrier carries those calls or an uptime SLA of its own, as at 18 October 2026, so this page does not claim one. Zian is in a partnership-application beta with no published prices; its home page lists a Free Forever plan (Intel + Widgets). If you would rather not run carriers or agents yourself, LeadsNow.ai is the done-for-you, pay-per-result option that runs on Zian.

What are the costs and SLAs for global voice networks?

A team running AI voice agents across countries pays a carrier per minute and carries the risk when calls stop connecting. The costs are easy to find and hard to compare; the SLAs are hard to find and easy to misread.

We read the SLA and legal pages of Twilio, Sinch, Plivo, Telnyx, Bandwidth and Infobip, plus each one’s pricing page, on 18 October 2026. Vonage’s legal and pricing pages returned an HTTP 403 bot challenge to every request we made that day, so Vonage is not scored: a gap in our reading, not a finding that Vonage lacks an SLA.

Three findings matter more than any single number:

• Only two of the six publish an availability percentage inside an SLA document. Twilio’s API SLA and Sinch’s Messaging, Voice and Verification SLA both say 99.95%.
• The SLA measures the platform, not the phone call. Twilio counts minutes its APIs are unavailable; Sinch counts minutes its service is not responding or not establishing outgoing connections. Neither addresses audio quality, post-dial delay or whether a German mobile network delivered your call.
• The carrier leg is excluded in writing. Twilio excludes “telecommunications provider-related problems or issues”. Sinch excludes failure or congestion “in any part of a network outside of where the Service are hosted”. Plivo’s support terms exclude “Telecommunication Carrier related issues”.

The SLA reading table: what each carrier commits to in writing

How we filled this table. We wrote these definitions before scoring any row, and applied them to every row the same way. A public SLA is a document on the vendor’s own site that states an availability percentage for a voice-capable service. A service credit is a stated remedy, as a percentage of fees, for missing that percentage. A marketing uptime figure is a percentage on a product page that does not appear in a contractual SLA document. We wrote “not found” only after reading the vendor’s legal index, its terms, its sitemap, its support or docs property, its pricing page and a product page, with a positive control on each saved file. Every cell was read on 18 October 2026.

Carrier	Public SLA document	Uptime commitment	What is measured	Credit at each breach tier	Shortest outage counted	Claim window and method
Twilio (APIs)	Twilio APIs Service Level Agreement, last updated 9 April 2026	99.95% a month; 99.99% for Twilio Enterprise Edition	Minutes the “application programming interfaces branded as ‘Twilio’” are unavailable	One tier: 10% of the month’s fees for the affected APIs, excluding taxes, carrier fees and surcharges, and third-party charges	5 minutes of continuous unavailability	Support ticket titled “SLA Claim” within 30 days of the last day of the month; times based on status.twilio.com; credit against future invoices, not refunds
Twilio (Flex)	Flex Critical Scenarios SLA, last updated 9 April 2026	99.99% a month	Six Flex contact-centre functions, including “Incoming Customer Voice Call” and “Outgoing Agent’s Voice Call”	5% below 99.99%; 15% below 99.0%; 100% below 95.0%, of the Flex access fees only	Counted in 5-minute intervals	Ticket within 30 days of month-end; credit only if over US$1
Sinch	Messaging, Voice, and Verification SLA, version 7.0, released 12 January 2024	“Reasonable endeavours” to a minimum 99.95% a month	Time the service “is not responding adequately to incoming requests, and/or are not establishing outgoing connections per region”	No credit schedule in the document	1 continuous minute	No credit claim process; faults reported via support.sinch.com with Call-IDs “within 7 days of issue”
Plivo	Platform Service Levels and Support Services (API), last updated 20 December 2024	No percentage in the document; a 99.99% “uptime promise” on its SIP trunking product page	The document defines support priorities, not availability	None stated	Not defined	Tickets via support.plivo.com; response-time targets sit on a support-plans URL that redirected to a contact-sales form when we read it
Telnyx	None found; terms say Telnyx “cannot guarantee the … availability or performance of the Services”	No contractual figure found; “99.99% uptime” on its WebRTC product page and “99.9% uptime” on several solutions pages	Not applicable	None found	Not applicable	None published; a Service Order controls where it contradicts the standard terms
Bandwidth	None found; its Communications Services Agreement refers to “any applicable service level agreement”	No contractual figure found; “99.999% core network uptime” on its home page	Not applicable	None found	Not applicable	None published; the agreement points to the Order Form for any support level other than standard
Infobip	None found; service terms say it cannot guarantee services “will be available to You at all times or free from faults or interruptions”	None found	Not applicable	None found	Not applicable	Planned maintenance announced at least 10 business days ahead
Vonage	Not assessed: vonage.com legal and pricing pages returned HTTP 403 to us on 18 October 2026

Read down the “what is measured” column: every published voice SLA we found measures whether the carrier’s platform answers, not whether your caller hears your agent.

Two caveats. The address telnyx.com/sla returned a Cloudflare block to us, but so did invented three-letter Telnyx paths such as telnyx.com/abc, while invented longer paths returned a plain 404, so the block is not evidence either way; the Internet Archive holds no capture of telnyx.com/sla, its captures under telnyx.com/legal/ include no SLA, and we found no SLA link in Telnyx’s 1,522-URL sitemap, its 1,034-article support sitemap, its terms or its pricing pages. Bandwidth’s agreement anticipates negotiated SLAs, and its Silver customer success plan advertises “SLA-backed responses and uptime”, but the plan’s PDF lists only support response times (P1: 30 minutes); we found no uptime SLA published on its legal index, its product-terms and supplemental-terms PDFs or its sitemap.

Marketing uptime is not an SLA

Three carriers publish uptime figures on product pages that match or beat the SLA documents above. Plivo’s SIP trunking page says “Redundant infrastructure across geographies and at least three local carrier connections per country back a 99.99% uptime promise”, and its Voice page shows “uptime 99.99 %”. Bandwidth’s home page shows “99.999% core network uptime”. Telnyx’s WebRTC page claims “99.99% uptime”, several of its solutions pages say “Rely on 99.9% uptime”, and its Voice AI Agents page promises “One platform, one SLA” without linking to an SLA document.

Those figures may well be accurate, but they are not in a document that defines downtime, exclusions, a remedy or a claim process. Plivo’s own SLA page, the one at plivo.com/legal/sla, contains no availability percentage and no credit, and lists “Telecommunication Carrier related issues” among the things Plivo has “no support obligation or liability” for.

The practical rule: an uptime figure without a measurement definition, an exclusions list and a remedy is a design goal, not a commitment. When a sales team quotes you 99.99%, ask which document it lives in, and ask for the order-form SLA if the answer is “none”.

What does a 99.95% SLA actually allow?

An availability percentage converts directly into minutes of permitted downtime per month. Both Twilio and Sinch measure by calendar month in minutes, so a 30-day month has 43,200 minutes and a 31-day month has 44,640.

Availability	Who uses this figure (and where)	Allowed downtime, 30-day month	Allowed downtime, 31-day month
99.9%	Reference point only	43.2 minutes	44.64 minutes
99.95%	Twilio APIs SLA; Sinch SLA	21.6 minutes	22.32 minutes
99.99%	Twilio Enterprise Edition and Flex SLAs; Plivo product-page figure	4.32 minutes	4.464 minutes
99.999%	Bandwidth home-page figure (not an SLA)	0.432 minutes (about 26 seconds)	0.446 minutes (about 27 seconds)

The formula is the same one Twilio prints: availability = (total minutes − unavailable minutes) ÷ total minutes × 100. So 21.6 minutes of downtime in a 30-day month is 43,178.4 ÷ 43,200 = 99.95%, exactly at the line.

The minimum-outage rule changes what counts. Twilio does not count any unavailability shorter than five continuous minutes. Four separate four-minute API failures in one month add up to 16 minutes of broken calls and zero minutes of SLA downtime. Sinch counts from one continuous minute, so the same four failures would count as 16 minutes against its 21.6-minute allowance.

A caller who hits dead air does not wait for minute five, which is why the failover design in how to build failover into an AI voice agent starts at the SIP trunk: the SLA will not pay for short gaps, so your routing has to absorb them.

A service credit, worked through with real terms

Here is the Twilio API SLA applied end to end. The inputs are illustrative; substitute your own invoice.

1. Your month. A 30-day month, 43,200 minutes. Your Twilio invoice is US$3,000, of which US$400 is carrier fees and surcharges and US$2,600 is fees for the Twilio APIs your agent used.
2. The incident. Twilio’s status page shows one 47-minute API outage, plus three separate four-minute blips.
3. What counts. The blips are each under five continuous minutes, so they are excluded. The 47 minutes count, unless the cause falls under an exclusion such as a telecommunications provider problem.
4. Availability. (43,200 − 47) ÷ 43,200 = 99.891%, which is below 99.95%.
5. The credit. 10% of the US$2,600 of API fees = US$260. Carrier fees and surcharges are excluded from the base, so the US$400 earns nothing.
6. The claim. Open a ticket at help.twilio.com titled “SLA Claim” within 30 days of the last day of the month, with the outage times from the status page and your documentation. The US$260 comes off a future invoice; it is not paid as a refund.

Twilio’s API credit has a single tier, so 99.9% and 90% availability earn the same 10%. And the credit is Twilio’s “sole and entire liability” for missing the threshold: US$260 is the whole remedy for those 47 minutes, whatever they cost you in missed bookings.

Under Sinch’s SLA the same incident earns no published credit at all, so the remedy is whatever your order form adds. On Telnyx, Bandwidth or Infobip you rely entirely on what you negotiate.

Published per-minute rates for a five-country basket

How we read the rates. Each figure is the carrier’s published pay-as-you-go list price in US dollars per minute, read on 18 October 2026, for its programmable voice or SIP product as named in the row. “Landline” is the carrier’s base row for the country; “mobile” is its base mobile row. Where a carrier prices the same destination by where the call starts, we give its default row and its same-country or EEA-origin row; surcharge rows for named high-cost origins (Twilio’s “Surcharged Zone” rows, Plivo’s “High Cost Origins” rows) are left out of the table and covered in the origin-pricing section below. Number rental and volume discounts are excluded. Coverage, meaning which countries each carrier serves at all, is a separate question, counted country by country in our voice API global coverage comparison.

Outbound, US$ per minute	Twilio Programmable Voice	Sinch Voice API (call from same-country number / any other origin)	Plivo Voice	Infobip Voice (average across networks)
United States, landline	0.0140 (US and Canada)	0.01 / 0.01	0.0115	0.0672
United Kingdom, landline	0.0158	0.01 / 0.65	0.2800 (0.0075 from EEA origins)	0.1148
United Kingdom, mobile	0.0305	0.022 / 0.65	0.3000 to 0.3200 by network (0.0270 from EEA origins)	Not split
Australia, landline	0.0252	0.02 / 0.0235	from 0.0230	0.0388
Australia, mobile	0.0750	0.03 / 0.0593	0.0730	Not split
India, landline	0.0699	0.0285 (one rate, any origin)	₹0.38, domestic calling within India only	0.0714
India, mobile	0.0496	0.0285 (one rate, any origin)	₹0.38, domestic calling within India only	Not split
Germany, landline	0.0283	0.03 / 0.381	1.0674 (0.0135 from EEA origins)	0.9210
Germany, mobile	0.3763 (0.0420 from EEA origins)	0.04 / 0.314	0.3850 (0.0359 to 0.0495 from EEA origins)	Not split

Two carriers publish US rates only on the pages we could read. Telnyx’s SIP trunking page shows US outbound local calls “starting at $0.005 per minute” and inbound local “starting at $0.0032”, with its Voice API adding “$0.002 per minute + the SIP Trunking fee”. Its country pages for the UK, Australia, India and Germany returned the same US content to us byte for byte, because the country rates load in the browser, so we have no Telnyx figures for the other four countries. Bandwidth’s pricing page shows US local outbound at $0.0100 and inbound at $0.0055 a minute, with “6-second billing increments”; its SIP trunking and international rates are by quote.

Inbound to a local number, US$ per minute	Twilio	Plivo	Infobip (average)	Telnyx	Bandwidth
United States	0.0085	0.0055	0.022	from 0.0032 (SIP), plus 0.002 on the Voice API	0.0055
United Kingdom	0.0100	0.0055	0.022	Not retrieved	Not on its pricing page
Australia	0.0100	0.0055	0.022	Not retrieved	Not on its pricing page
India	No inbound number rate on its India page	₹0.38, domestic only	Not listed	Not retrieved	Not on its pricing page
Germany	0.0100	Not supported	0.022	Not retrieved	Not on its pricing page

Sinch’s inbound rates render in the browser and were not in our saved page, so Sinch has no inbound column. Infobip’s figures are averages: its pricing page says “We display the average price across all supported networks for each country. Per-network pricing is available in Portal.” So its Germany figure, US$0.92095, is a blend across networks, not a landline rate, and it sits nearer the default rates other carriers list for Germany, such as Plivo’s US$1.0674 to landlines, than their EEA-origin rates.

The tables support one plain statement: for the same destination, the cheapest and dearest published rates differ by more than ten times in the US, the UK and Germany, so one blended rate for a multi-country agent hides most of the cost. Every published US outbound rate we read sits between US$0.005 and US$0.0672 a minute. UK landline rates run from US$0.0075 to US$0.65 and German landline rates from US$0.0135 to US$1.0674, the low end from an EEA origin and the high end from other origins. Australia and India are tighter: US$0.02 to US$0.0388 and US$0.0285 to US$0.0714 for landlines.

Why the same call can cost nearly 30 times more depending on where it starts

Origin-based pricing is the cost line multi-country teams most often miss. Sinch’s price list, valid from 15 October 2026, carries an origination-prefix column. A call to a UK mobile costs US$0.022 a minute where the origination prefix is 44, and US$0.65 from any other origin. Take 10,000 minutes a month to UK mobiles:

• Sinch, call originating from a +44 prefix: 10,000 × 0.022 = US$220.
• Sinch, any other origin: 10,000 × 0.65 = US$6,500, about 29.5 times as much.
• Twilio’s base UK mobile row: 10,000 × 0.0305 = US$305. Twilio also lists higher UK rows for calls from its “Surcharged Zone” origination prefixes, at US$0.4730 a minute to UK mobiles.

Germany shows the same shape. Sinch charges US$0.03 to a German landline from a 49 prefix and US$0.381 from elsewhere, 12.7 times as much. Twilio charges US$0.3763 to German mobiles by default and US$0.0420 from EEA origination prefixes. Plivo’s published rows show the widest spread: US$0.0135 a minute to German landlines from EEA origins against US$1.0674 by default, and UK landline rows from US$0.0075 from EEA origins up to US$6.0000 from what Plivo calls “Ultra High Cost Origins”. Bandwidth’s outbound product terms say the same thing in contract language: “pricing for some destinations may vary based on the origin of the Outbound Call.”

For an AI agent, the caller ID you present is a pricing decision as well as a trust decision. Check the origin column in each carrier’s rate file before you model a multi-country bill. Model and speech costs sit on top of every carrier rate here; our page on voice AI cost per minute breaks those down.

Billing increments move the total too: Sinch’s price list shows 60-second initial and additional rounding for these five countries, Bandwidth states 6-second increments for its US Voice API, and Plivo’s India page says “Pulse is 30 seconds”. A 61-second call is two billed minutes on a 60/60 schedule and 66 seconds on a 6-second schedule.

How to claim a service credit, and what to log before you need one

The claim processes we found are short and dated from month-end, so the work is in the logging you do beforehand.

1. Subscribe to the status page. Twilio’s claim requirements ask for the dates and times of unavailability “calculated based on the status page” at status.twilio.com. Sinch announces scheduled service windows on status.sinch.com at least 10 business days ahead, and Infobip announces planned maintenance at least 10 business days ahead on its status page.
2. Log your own timestamps. Record when calls stopped connecting and recovered, by region. That is the “documentation” Twilio asks for, and it shows whether the outage reached the five-minute floor.
3. Keep call identifiers for a week. Sinch’s fault report asks for relevant Call-IDs “within 7 days of issue”, plus sample numbers or CDRs.
4. Diagnose the layer. If the failure was a downstream mobile network or an internet path outside the carrier, it is excluded under Twilio’s and Sinch’s SLAs. Know which before you file.
5. File inside the window. Twilio requires the claim within 30 days of the last day of the month, with “SLA Claim” as the ticket subject. Flex claims use “Flex Critical Scenarios SLA Claim”.

Twilio’s API SLA lists seven exclusions, and a claim fails if the outage falls under any of them: (a) factors outside Twilio’s reasonable control, including telecommunications provider problems and internet problems beyond where Twilio controls access; (b) action or inaction of the customer or a third party, other than Twilio’s agents and subcontractors; (c) the customer’s or a third party’s equipment, software or technology, other than equipment in Twilio’s direct control; (d) scheduled maintenance with at least 24 hours’ notice; (e) emergency maintenance, whatever the notice; (f) alpha, beta, limited-release or developer-preview APIs; and (g) unavailability shorter than five continuous minutes. The API SLA also does not apply to products no longer available or supported, or to Segment, Stytch or the Flex Critical Scenarios, which have their own terms.

Sinch’s list has five: scheduled maintenance with prior notice; failure or congestion in a network outside where Sinch hosts the service; causes beyond either party’s reasonable control; suspension under the agreement; and customer-caused issues such as not providing requested information or not interfacing as Sinch instructs.

When the SLA answer changes

The public documents are the floor. The answer moves in five situations:

• You buy an enterprise edition. Twilio’s threshold rises from 99.95% to 99.99% in months when you hold Twilio Enterprise Edition, or Administration Edition bought before 24 October 2024.
• You sign an order form. Sinch’s SLA applies to services “ordered by the Customer under an Order Form”. Telnyx’s terms say a Service Order controls where it contradicts the standard terms. Bandwidth’s agreement refers to “any applicable service level agreement”. On these carriers, the SLA you get is the one you negotiate.
• You use a beta product. Twilio excludes anything “identified as alpha, beta, not generally available, limited release, developer preview”. Preview labels are common on voice-AI features: Sinch’s voice pricing page labels Voice Relay, Streams and AI text to speech “Public Preview”. Check each carrier’s labels and terms before you rely on a preview feature.
• You run a contact centre on Flex. Twilio’s Flex SLA is the only published SLA we found that names call outcomes, such as “Incoming Customer Voice Call”, at 99.99% with tiered credits on Flex access fees.
• Your traffic is SIP trunking only. Twilio’s API SLA covers “application programming interfaces branded as ‘Twilio’” and does not name Elastic SIP Trunking or call media. We would not assume a trunk-only deployment is covered without asking Twilio in writing.

For a US-only build, the bill-level comparison in Twilio vs Telnyx for AI voice agents prices a 100,000-minute month on both carriers.

Frequently asked questions

Does Twilio have an SLA for voice calls?

Twilio’s Twilio APIs Service Level Agreement commits to 99.95% monthly availability for APIs branded “Twilio”, with a 10% credit if it is missed. It measures API unavailability, not call quality, and excludes telecommunications provider problems. Its separate Flex SLA covers voice calls inside Twilio Flex at 99.99%.

How much downtime does 99.95% uptime allow?

In a 30-day month, 99.95% allows 21.6 minutes of downtime; in a 31-day month, 22.32 minutes. Twilio does not count outages shorter than five continuous minutes, while Sinch counts from one continuous minute.

Do CPaaS SLAs cover outages on the mobile carrier side?

Not in the documents we read on 18 October 2026. Twilio excludes “telecommunications provider-related problems or issues”, Sinch excludes failure or congestion in networks outside where its service is hosted, and Plivo’s support terms exclude “Telecommunication Carrier related issues”.

Why do some carriers charge more for the same destination?

Origin-based pricing. Sinch’s price list charges US$0.022 a minute to UK mobiles where the origination prefix is 44 and US$0.65 from any other origin. Twilio charges US$0.3763 to German mobiles by default and US$0.0420 from EEA origination prefixes.

Which carriers publish no uptime SLA?

We found no public uptime SLA for Telnyx, Bandwidth or Infobip on their legal, support and pricing pages as at 18 October 2026, although one Telnyx address was blocked to us. Telnyx and Bandwidth terms point to order forms. Plivo’s SLA page covers support only. Plivo, Telnyx and Bandwidth all state uptime figures, from 99.9% to 99.999%, on product pages rather than in an SLA document. We could not load Vonage’s pages.

Does Zian publish an uptime SLA or name its carrier?

No. As at 18 October 2026, zian.ai does not publish an uptime SLA or name the carrier under its calls. Zian is in a partnership-application beta, so ask during your application and check the carrier terms on this page separately.

Where every figure on this page comes from

Figure	Who published it	Link	Date read
99.95% and 99.99% API thresholds; 10% credit; credit excludes carrier fees; 30-day claim; “SLA Claim” subject; 5-minute floor; exclusions (a)–(g); not refunds; sole remedy	Twilio	Twilio APIs SLA	18 October 2026
List of Twilio SLAs and covered services; last updated 9 April 2026	Twilio	Service Level Agreements	18 October 2026
Flex 99.99%; credits 5%, 15%, 100%; 5-minute intervals; US$1 minimum	Twilio	Flex Critical Scenarios SLA	18 October 2026
Twilio outbound rates (US, UK, AU, IN, DE; EEA and Surcharged Zone rows)	Twilio	Outbound voice pricing CSV	18 October 2026
Twilio inbound local rates; no India inbound number rate	Twilio	US, UK, Australia, India, Germany voice pricing	18 October 2026
Sinch 99.95% “reasonable endeavours”; 1-minute unavailability; five exclusions; 10 business days notice; Call-IDs within 7 days; version 7.0	Sinch	Messaging, Voice, and Verification SLA	18 October 2026
Sinch outbound rates by origin; valid from 15 October 2026; 60/60 rounding	Sinch	Voice API price list, linked from sinch.com/voice/pricing	18 October 2026
Plivo SLA page contents; carrier-issue exception; last updated 20 December 2024	Plivo	Service Level and Support	18 October 2026
Plivo “99.99% uptime promise”	Plivo	SIP trunking, Voice	18 October 2026
Plivo rates (US, UK, AU, IN, DE), including EEA-origin, default and “Ultra High Cost Origins” rows; India 30-second pulse; Germany inbound not supported	Plivo	US, UK, Australia, India, Germany voice pricing	18 October 2026
Telnyx “cannot guarantee” availability (11.11); availability “not guaranteed” (18.7); Service Order precedence	Telnyx	Terms and Conditions of Service	18 October 2026
Telnyx US SIP rates from US$0.005 out, US$0.0032 in; Voice API US$0.002	Telnyx	SIP trunking pricing, Voice API pricing	18 October 2026
“99.99% uptime” (WebRTC); “Rely on 99.9% uptime” (solutions pages); “One platform, one SLA”	Telnyx	WebRTC, healthcare solutions, Voice AI Agents	18 October 2026
1,522-URL site sitemap; 1,034-article support sitemap	Telnyx	telnyx.com sitemap, support sitemap	18 October 2026
“any applicable service level agreement”	Bandwidth	Communications Services Agreement	18 October 2026
Origin-based pricing clause; product-terms PDFs	Bandwidth	Product terms	18 October 2026
“99.999% core network uptime”	Bandwidth	bandwidth.com	18 October 2026
Silver plan “SLA-backed responses and uptime”; P1 response 30 minutes	Bandwidth	Customer success plans, plan PDF	18 October 2026
US rates US$0.0100 out, US$0.0055 in; 6-second increments	Bandwidth	Pricing	18 October 2026
Infobip “cannot … guarantee” availability (4.2); last updated 29 September 2026	Infobip	Service terms and conditions	18 October 2026
Planned maintenance notice of 10 business days	Infobip	Incident management	18 October 2026
Infobip average outbound and inbound rates; averaging disclaimer	Infobip	Voice pricing	18 October 2026
Downtime minutes per availability level; credit and origin-pricing worked examples	Zian (arithmetic from the rows above)	This page	18 October 2026