Start with Telnyx if you need low-cost programmable calls on a private network. Sinch is the stronger choice for enterprise teams already on Sinch messaging, while ElevenLabs suits voice agents with natural voices. The other 3 tools in the list cover narrower or pricier cases.
We run no ads and no affiliate links at launch. The order of a list follows our published method.
Prices are the starting prices vendors list publicly. Plans change, so confirm the current price on the vendor page before you buy.
Key takeaways
This list starts with Telnyx, which is built for low-cost programmable calls on a private network.
A fair warning about ElevenLabs: language model and telephony bill separately.
ElevenLabs has the lowest monthly starting price in this list: $6 per month.
Bandwidth fits US voice on an owned carrier network.
ElevenLabs stands alone as the one tool in this list with a free plan.
Tool
Best for
Standout feature
Listed starting price
Free plan or trial
1 Telnyx
Low-cost programmable calls on a private network
Low blended US outbound rate
$0.007 per minute outbound US (Voice API plus SIP trunking)
No free plan
2 Sinch
Enterprise teams already on Sinch messaging
Broad messaging
$0.01 per minute outbound US (pay as you go)
Free trial (60 days)
3 ElevenLabs
Voice agents with natural voices
Large voice library with cloning
$6 per month (Starter, 75 agent minutes included)
Free plan
4 Twilio
Most teams building programmable calls
US local numbers at $1.15 a month
$0.0140 per minute outbound US local (pay as you go)
Free trial
5 Deepgram
One API for speech, model and voice
Standard agent bundles speech
$0.075 per minute Standard Voice Agent (pay as you go)
Free trial
6 Bandwidth
US voice on an owned carrier network
US local numbers at $0.30 a month
$0.0100 per minute outbound US local (pay as you go)
Free trial
How we picked
Each tool was reviewed on what its vendor publishes: price, plan limits, core features and stated drawbacks. We did not run hands-on tests. Where a vendor does not publish a price, the entry says quote. Where a price depends on seats or billing term, the entry says so. We ordered the tools by general fit, then by how clearly pricing is shown, then by free access. A buyer with unusual needs may reorder them.
01
Telnyx
Best for: Low-cost programmable calls on a private network
Choose Telnyx for low-cost programmable calls on a private network. Telnyx splits a call into two charges, and that split is the fact most roundups miss. More specifically, the Voice API system fee is $0.002 per minute and US SIP trunking brings from $0.005 per minute outbound and from $0.0032 per minute inbound.
So an US outbound Voice API call totals about $0.007 per minute. For everyday use, there is no monthly system fee on pay as you go. Also, one limit: the voice engine rounds each call up to 60-second increments, so short calls cost more than the per-minute rate recommends. Telnyx Voice AI Agents cost $0.05 per minute for orchestration, speech to text and text to speech, with language model tokens billed separately, and T The starting price for Telnyx: $0.007 per minute outbound US. Telnyx does not include a free plan.
The main drawbacks: AI engine rounds to 60-second increments.
Pros
Low blended US outbound rate
Voice AI engine at $0.05 per minute with speech included
Cons
Voice API fee and SIP rate bill as two lines, which hides the true rate on a first read
Price: $0.007 per minute outbound US (Voice API plus SIP trunking), as listed by the vendor (Telnyx website). Confirm the current plan price before you buy.
Skip it if this limit matters to you: voice API fee and SIP rate bill as two lines, which hides the true rate on a first read.
Verdict. A sensible pick for low-cost programmable calls on a private network.
02
Sinch
Best for: Enterprise teams already on Sinch messaging
Think of Sinch for enterprise teams already on Sinch messaging. Sinch Voice API lists domestic outbound calls from $0.01 per minute, with phone to phone inbound at $0.004 per minute and outbound at $0.01 per minute in its SMB pricing, and numbers from $0.80 per month. In daily use, app to app calling is free for the first 2,500 minutes per month and then $0.002 per minute, whereas SIP connectivity costs $0.0025 per minute.
Trade-offs to know: Newer AI voice products need a quote.
The newer AI layer is the weak point for self-serve buyers. Also, one limit: Sinch bills voice by the full minute, so a 20-second call costs the same as a 60-second call. As for the details, Sinch also sells SMS, email, RCS and verification from the same platform. More specifically, voice Relay and Agentic Conversations need a workload-specific quote, and Sinch labels Voice API v2 as preview, telling production projects to use v1 What you pay to start with Sinch: $0.01 per minute outbound US. Trial and free access: Free trial (60 days).
Pros
Broad messaging
Voice and verification range
Cons
Voice API v2 is labelled preview, so production work should use v1
Price: $0.01 per minute outbound US (pay as you go), as listed by the vendor (Sinch website). Confirm the current plan price before you buy.
Skip it if this limit matters to you: voice API v2 is labelled preview, so production work should use v1.
Verdict. A sensible pick for enterprise teams already on Sinch messaging.
03
ElevenLabs
Best for: Voice agents with natural voices
Pick ElevenLabs if this describes you: voice agents with natural voices. The plans run from Free at $0 with 15 agent minutes and 4 concurrent calls, through Starter at $6 per month with 75 minutes, Creator at $22 with 275 minutes, Pro at $99 with 1,238 minutes, Scale at $299 with 3,738 minutes, and Business at $990 with 12,375 minutes and 40 concurrent calls.
One limit: burst pricing doubles the minute when you pass your concurrency cap, which can surprise teams during a traffic spike. Price to start with ElevenLabs: $6 per month. ElevenLabs includes a free tier.
Limits to weigh: concurrency caps rise only with higher plans.
Pros
Large voice library with cloning
Agent minutes at $0.08 per minute on every plan
Cons
Language model and telephony bill separately
Price: $6 per month (Starter, 75 agent minutes included), as listed by the vendor (ElevenLabs website). Confirm the current plan price before you buy.
Skip it if this limit matters to you: language model and telephony bill separately.
Verdict. A sensible pick for voice agents with natural voices.
04
Twilio
Best for: Most teams building programmable calls
Twilio is built with most teams building programmable calls in mind. The trade-off is that a straightforward AI call can reach about $0.084 per minute before model costs, which is more than the bundled agent rates from Deepgram Twilio Programmable Voice charges $0.0140 per minute for US local outbound calls and $0.0085 per minute for inbound calls on a local number, with toll-free inbound at $0.0220 per minute.
Limits to weigh: volume discounts start only at higher usage.
For everyday use, ConversationRelay layers on $0.07 per minute on top of telephony when you link a call to your own language model, and Media Streams brings $0.0044 per minute for raw audio. Also, an US local number costs $1.15 a month and recording layers on $0.0025 per minute. The starting price for Twilio: $0.0140 per minute outbound US local. Free access: Free trial.
Pros
US local numbers at $1.15 a month
ConversationRelay AI bridge at $0.07 per minute
Cons
AI features bill as separate extras
Price: $0.0140 per minute outbound US local (pay as you go), as listed by the vendor (Twilio website). Confirm the current plan price before you buy.
Skip it if this limit matters to you: AI features bill as separate extras.
Verdict. A sensible pick for most teams building programmable calls.
05
Deepgram
Best for: One API for speech, model and voice
Deepgram targets one API for speech, model and voice. The Standard level costs $0.075 per minute on pay as you go, or $0.068 on the Growth plan, which needs a commitment from $4,000 per year. More specifically, one limit: a phone call needs a separate carrier like Twilio.
So the all-in minute is about $0.0895 with Twilio outbound and Media Streams. Speech to text alone costs $0.0077 per minute for Nova-3 streaming on pay as you go, and $0.0043 per minute for pre-recorded Nova-3 audio. Deepgram starts at: $0.075 per minute Standard Voice Agent. Trial and free access: Free trial.
Limits to weigh: Advanced level jumps to $0.163 per minute.
Pros
Standard agent bundles speech
Bring-your-own levels from $0.05 per minute
Cons
Telephony is not included, so you add a carrier bill
Price: $0.075 per minute Standard Voice Agent (pay as you go), as listed by the vendor (Deepgram website). Confirm the current plan price before you buy.
Skip it if this limit matters to you: telephony is not included, so you add a carrier bill.
Verdict. A sensible pick for one API for speech, model and voice.
06
Bandwidth
Best for: US voice on an owned carrier network
Bandwidth was built for US voice on an owned carrier network. Bandwidth Voice API lists US local outbound calls at $0.0100 per minute and US local inbound calls at $0.0055 per minute, with 6-second billing increments. So a 12-second call bills for 12 seconds instead of a full minute.
Watch-outs: number rental is not shown for every number type on the pricing page.
Standard text to speech costs $0.0007 per 100 characters, recording costs $0.0020 per minute, and answering machine detection costs $0.0060 per call. An US local number is listed at $0.30 a month on the voice page. In practice, Bandwidth owns and runs its own nationwide all-IP voice network, with number inventory in about 70 countries. As for the details, one limit: the pricing page notes that full product pricing is available on request and that rates exclude taxes and fees, so some services need a sal Bandwidth starts at: $0.0100 per minute outbound US local. On free options: Free trial.
Pros
US local numbers at $0.30 a month
US local outbound at $0.0100 per minute
Cons
Full product pricing is available on request for some services
Price: $0.0100 per minute outbound US local (pay as you go), as listed by the vendor (Bandwidth website). Confirm the current plan price before you buy.
Skip it if this limit matters to you: full product pricing is available on request for some services.
Verdict. A sensible pick for US voice on an owned carrier network.
What to check before you buy
Prices are a starting point; ElevenLabs shows $6 per month to start, so verify what your seat count would cost.
Test the free plan from ElevenLabs against your real workload before relying on it.
Take the trial from Sinch, Twilio, Deepgram and Bandwidth and push a live task through it before you pay.
Read the limits listed for Telnyx first: voice API fee and SIP rate bill as two lines, which hides the true rate on a first read.
Before rolling out Twilio, verify the integrations your workflow depends on.
Confirm minimum seats and billing terms with Sinch before you commit to an annual plan.
FAQ
What is the cheapest option for voice API providers in this list?
On published monthly entry prices, ElevenLabs is the lowest here at $6 per month. Entry price is only part of the cost, because seats, add-ons and billing terms can change the total.
Which is the best voice API provider for AI voice agents?
For most readers this guide would start with Telnyx, built for low-cost programmable calls on a private network. Sinch is the alternative for enterprise teams already on Sinch messaging, and ElevenLabs for voice agents with natural voices. The right answer depends on budget and workflow.
Which options for voice API providers have a free plan?
ElevenLabs is the only tool in this list with a free plan. Free plans usually cap users, projects or features, so check the limits on the vendor page. Several others offer a trial instead.
Do these tools publish their prices?
Yes. Every tool from Telnyx to Bandwidth shows a starting price, though seats, add-ons and billing terms still change the final cost.
How are the tools for voice API providers ordered?
Telnyx sits first and Bandwidth last because the order follows typical fit for voice API providers, then price clarity and free access. It is a judgement call, so read each profile for fit and limits.