ElevenLabs publishes five monthly prices: $6, $22, $99, $299 and $990. Annual billing is badged as two months free and checks out exactly, at $5, $18.33, $82.50, $249.17 and $825 a month.
None of those is a price for a voice agent minute, because ElevenLabs does not sell minutes.
It sells characters, metered as credits, and a voice agent budget is written in minutes of call. The conversion is the entire subject of this page, and it is the step almost every comparison skips. Get it wrong in the cheap direction and your synthesis line comes in at six times what you modelled.
What follows: the plan ladder, the conversion from characters to call minutes, where that lands against the cheap alternative, the vendor's own agent product undercutting its own voice, what a premium voice does to a whole stack, and why this vendor carries no computed all-in figure on this site.
What does ElevenLabs cost?
$6 a month at the bottom and $990 at the top, with a free tier underneath both. The same subscription covers the creative tools, the API and the agents.
| Plan | Billed monthly | Billed annually | Effective annual saving |
|---|---|---|---|
| Free | $0 | $0 | n/a |
| Starter | $6 | $5.00 | 16.7% [computed] |
| Creator | $22 | $18.33 | 16.7% [computed] |
| Pro | $99 | $82.50 | 16.7% [computed] |
| Scale | $299 | $249.17 | 16.7% [computed] |
| Business | $990 | $825.00 | 16.7% [computed] |
Two months free on a twelve-month term is a sixth off, and the badge and the arithmetic agree to the cent on all five tiers. That is worth saying plainly, because this site has found vendors whose annual badge and annual price do not match.
What the plan card does not tell you is what a minute costs.
The meter behind the plan is characters, not time
Every plan buys an allowance of credits, and credits are consumed by characters of text turned into audio.
A meter denominated in characters is the right meter for the vendor, because characters are what the model actually processes. It is the wrong meter for a buyer costing a phone call, who knows how many minutes of conversation they expect and has no idea how many characters that is.
Two conversions sit between the plan card and your budget, and each one is a place to go wrong. The first turns credits into minutes of generated audio. The second turns minutes of generated audio into minutes of call, because the agent speaks roughly half of a two-party conversation.
Skip the second and you halve your own estimate by accident, in the vendor's favour.
What one minute of call actually costs
| Tier | Per minute of generated audio | Per minute of call at 50% talk | Against the cheap option |
|---|---|---|---|
| Creator | $0.18 | $0.090 [computed] | 6.4x [computed] |
| Pro | ~$0.17 | $0.085 [computed] | 6.1x [computed] |
| Cartesia Sonic, Scale, for comparison | $0.028 | $0.014 | 1x |
So a premium ElevenLabs voice is roughly six times the cheapest credible synthesis in the category, per minute of call, at list.
That spread is the widest of any component in a voice agent stack. Recognition varies by tenths of a cent between vendors and telephony is a carrier rate you cannot argue with. Synthesis is where the money actually moves, and it is the decision teams make last, in a settings panel, without the pricing spreadsheet open.
This site's own note about that spread is wrong
The vendor's own agent product undercuts its own voice
ElevenLabs Agents charges $0.08 a call minute at every paid tier, and that rate includes both the synthesis and the speech recognition.
Standalone synthesis on Creator is $0.090 a minute of call, and includes nothing else.
Buying the whole agent from ElevenLabs is cheaper than buying only the voice from ElevenLabs, by about a cent a minute, or 11% [computed]. That inverts the pattern this site was built to expose, where a platform layer marks up components you could have bought directly. Here the platform is where the discount lives.
If you are building an agent and had planned to bring ElevenLabs synthesis into a third-party orchestrator, price the vendor's own agent product first. It may be the cheaper route to the same voice.
What a premium voice does to a whole stack
The clearest published demonstration is not on ElevenLabs' own pages. It is on LiveKit's component rate card, which prices an agent session at a fixed $0.0100 a minute and lets you fill in every other layer from a menu.
On that card, text to speech runs from free (Deepgram Flux) to $0.1800 a minute for ElevenLabs Eleven v3, read 25 August 2026. Assemble the cheapest credible option at every layer and the total is $0.0212 a minute. Assemble the dearest and it is $0.2496. That is 11.8 times at one advertised rate, and the synthesis line alone accounts for 79% of the difference.
This site works through that spread in the full costing rather than here.
Against the $0.0140 this cost model carries for synthesis, Eleven v3 at $0.1800 a minute is 12.9 times.
Third parties quote a third rate again
Vapi's usage calculator, read 6 August 2026, prices ElevenLabs synthesis at $0.036 a minute inside a voice agent stack.
That sits between this site's cheap reference and its premium one, and it is a fourth number for the same vendor's audio. Flash-class models sit in that band, and a platform's default selection is usually a faster and cheaper model than the flagship a buyer auditions on the marketing page.
The practical instruction is short. Find out which ElevenLabs model your platform has selected by default, because the name of the vendor tells you almost nothing about the rate.
Why this vendor has no all-in figure on this site
Six entities in this directory carry a computed all-in per-minute floor. ElevenLabs is not one of them, and that is deliberate rather than an omission.
A cost stack exists to answer one question: what does the advertised rate leave out. ElevenLabs the synthesis product does not advertise a rate for a voice minute at all, so there is nothing to add components to. It is one of the four components, priced on its own meter.
The vendor's agent product does carry a stack, at $0.08 advertised and $0.1048 computed, a multiple of 1.31.
When the premium voice is worth six times the cheap one
When a customer listens for more than a few seconds and prosody carries the experience. Consumer apps, media, anything where a synthetic-sounding agent loses the user before the second sentence.
When you are cloning a specific voice, which the cheaper vendors do less well or not at all.
And not when your agent reads back a reference number. On a phone call band-limited to 8kHz, the difference between a premium voice and a good cheap one is frequently inaudible, and you are paying six times for a quality the network is throwing away. Generate one minute of your actual script on both and play it down a real phone line before you decide.
What the vendor's own model calculator reveals
ElevenLabs publishes something almost nobody else in this category does: a per-minute cost for each language model, converted from token pricing by the vendor itself, sitting behind a dropdown on its agents pricing page.
Read on 5 August 2026 it spanned $0.0005 a minute for GPT-5 Nano to $0.0446 for GPT-5.5. That single dropdown moves the all-in floor of an agent minute by about 47%, which is more than any tier decision on the page, and a custom model of your own is billed at zero.
Worth knowing where the vendor's assumptions differ from this site's. Working backwards from that calculator, it appears to assume roughly 1,240 input tokens a turn where this site's cost model assumes 1,800. Their figure is the more optimistic of the two, and long calls with growing transcripts push the two further apart.
How to price your own synthesis line in ten minutes
- Take your expected minutes of call, not minutes of audio. Halve for the talk ratio, or measure your own from a recording if you have one, because agent-led calls skew above half.
- Convert the plan allowance into minutes of audio using the vendor's own conversion rather than a character estimate of your own.
- Multiply out at both ends of the spread, $0.014 and $0.085 per minute of call, and see whether the difference changes your decision. Frequently it changes it more than the platform choice did.
- Audition on a phone line, not on a laptop. The comparison that matters happens at 8kHz.
- Price the vendor's own agent product as an alternative, because at $0.08 a minute with transcription included it undercuts the standalone voice.
What this page does not know
Nine synthesis routes, priced, with the day each was read
The same company's audio appears at four different numbers in this table, which is the reason it exists. Convert every row to a minute of call before comparing anything, and halve any figure denominated in generated audio.
| Synthesis route | Published rate | Per minute of call | What it includes | Read |
|---|---|---|---|---|
| ElevenLabs Creator | $0.18 per min of generated audio | $0.090 [computed] | Synthesis only | 2026-08-05 |
| ElevenLabs Pro | ~$0.17 per min of generated audio | $0.085 [computed] | Synthesis only | 2026-08-05 |
| ElevenLabs Eleven v3, on LiveKit's card | $0.1800 a minute | Not stated as a call minute on the card | Synthesis only | 2026-08-25 |
| ElevenLabs, as Vapi's calculator selects it | $0.036 a minute | As the calculator states it | A platform's default model, not the flagship | 2026-08-06 |
| ElevenLabs Agents | $0.08 per minute of call | $0.08 | Synthesis and speech recognition | 2026-08-05 |
| Cartesia Sonic, Scale | $0.028 per min of generated audio | $0.014 [computed] | Synthesis only | 2026-08-07 |
| Deepgram Flux TTS, on LiveKit's card | Free | $0.00 | Free inside Deepgram Voice Agent through 12 Sep 2026 | 2026-08-25 |
| Fish Audio s2.1-pro | $15.00 per million UTF-8 bytes | Not computable from bytes | Hosted API alongside open weights | 2026-08-07 |
| Kokoro | $0, open weights | Your compute | Local deployment, no hosted service | 2026-08-07 |
Rows one, three and four are all ElevenLabs, at $0.18, $0.1800 and $0.036 a minute. The third is the flagship model on somebody else's rate card and the fourth is a faster, cheaper model a platform picked for you in a settings panel. Naming the vendor tells you almost nothing about the rate, which is why the practical instruction on this page is to find out which model your platform selected by default.
Row five is the one that inverts the usual pattern. The vendor's own agent product bundles the same synthesis with speech recognition at $0.08 a minute of call, which undercuts its own standalone voice on Creator by about a cent a minute, or 11% [computed]. The platform layer is where the discount lives rather than where the markup is.
What breaks in month three
The talk ratio you assumed. Every per-call-minute figure on this page halves a per-audio-minute rate on the assumption that the agent speaks half the call. Agent-led calls skew above half, and a script that reads back an address or a reference number can skew well above it. That single input moves your synthesis line more than any tier decision on the pricing page.
The model somebody changed in a dropdown. The vendor's own model calculator, read 5 August 2026, spanned $0.0005 a minute for GPT-5 Nano to $0.0446 for GPT-5.5, which moves the all-in floor of an agent minute by about 47%. Nobody reviews that setting after launch, and it is not on the invoice line anybody looks at.
And the allowance runs out mid-month. The plan buys credits, credits are consumed by characters, and characters are not a unit anyone budgets in. A team that modelled minutes and bought credits finds out which of the two the vendor is counting somewhere around week three of the first busy month.