ElevenLabs Pricing 2026: Cost Per Finished Minute by Tier

Source: https://elevenlabs.io/pricing and https://elevenlabs.io/pricing/api, both verified 2026-09-29.

Now the disagreement, named. bigvu.tv, in a post dated 2026-03-24, states Starter at $5 and Business at $1,320. websitevoice.com, dated 2025-05-29, carries the same $5 entry and the same $1,320 ceiling, along with a $330 Scale tier. cekura.ai is the strangest case of the three, because it contradicts itself inside one page: its overview table shows the stale $5 and $1,320 figures while its own FAQ text and its SERP snippet carry the current $6 and $990. flexprice.io does the same trick in the opposite direction, correct in its search snippet, stale in its body. The staleness is not a vendor fork. It is one repricing in 2026 that three sources never followed, which is also why their mid-tier credit counts (100k, 500k, 2M, 11M) are wrong by roughly the same margin.

One page in this result set is accurate and recent: felloai.com, updated 2026-09-23, carries the correct ladder and a correct cost per minute column. That is worth saying plainly, because it means a current ladder is no longer a reason to read a page. The arithmetic below is.

Two more vendor-side facts worth holding before the math. The Starter, Scale and Business numerals do not render in a plain text extract of the plan cards; they are confirmed on the vendor's API page, which prices the API on the same six subscription tiers rather than on separate API subscriptions. And the vendor's own credit allocation is not internally consistent: the subscription page sells Starter at 30,000 credits, while the API page's Starter cell reads 273,000 characters of TTS under the v4 model block. Same vendor, same day, a factor of nine. The vendor explains neither figure on either page, so do not let a single blog post convince you it is settled.

What you are actually buying: credits, characters and the ElevenLabs AI voice generator

One credit equals one character of input text, and other operations are charged per second of audio processed. That is the vendor's own definition, on https://elevenlabs.io/docs/overview/intro, read 2026-09-29.

The conversion you need is also published, and this is the part almost nobody cites. 1,000 credits is about 1 minute of finished narration, and it holds at every tier on the vendor's own compare row: 10,000 credits to ~10 minutes, 30,000 to ~30, 121,000 to ~121, 600,000 to ~600, 1.8 million to ~1,800, 6 million to ~6,000. The vendor's API page pairs the same conversion in its rate card, listing 1,000,000 characters beside ~1,000 minutes and 4,500,000 beside ~4,500. That is four independent vendor sources agreeing on one conversion, and the only honest caveat is the vendor's own tilde. Included minutes are approximate, and the vendor says so on the page.

If you have been reading character allowances anywhere and wondering why the plan names changed, the help centre answers it directly: "Credits were previously referred to as 'characters.' We updated this terminology to streamline pricing and make it easier to understand, though the value remains unchanged." Half the explainers in this result set still quote the old word, and a few still quote the old numbers with it.

Two multipliers that move your real cost

The model you pick halves or doubles the bill. The vendor lists Eleven Flash v2.5 as a faster model at 50% lower price per character for API generations. Half the credits per character means roughly 500 credits per finished minute instead of 1,000. The vendor's own API table confirms the doubling arithmetically: its Flash block carries exactly twice the included characters of the multilingual block at every tier. On the standard multilingual models you are buying minutes at the advertised rate. On Flash you are buying roughly two minutes per thousand credits, and the audio sounds different, which is a decision about your brand, not only about your budget.

Where ElevenLabs text to speech fits a marketing workflow

Here is how the volume usually arrives. One project needed a voiceover and the plan looked cheap. Then the second project shows up, then the product demo, then the twelve social cutdowns of the same script. Retakes are where the arithmetic bites, because you are charged per generation rather than per download, and a regeneration multiplies the cost of that minute. oakgen.ai, in a post dated 2026-08-13, supplies the cleanest framing available for this: credits per accepted piece equals generations times attempts per accepted piece, and it is right to warn against borrowing anyone else's waste rate. What follows is a labelled example, not a universal figure. At 1.25 attempts per accepted minute your effective cost rises by 25%. At two attempts per accepted minute it doubles. Import your own retake rate, not mine.

ElevenLabs pricing per finished minute, tier by tier

Everything below is derived, and the derivation is one division: monthly price divided by vendor-published included minutes. Nothing here is invented, and the cross-check is in the last column.

The left two columns are the vendor's published rows. The derived column agrees with the vendor's extra-minute row at every single tier, which is what makes it safe to publish: two vendor columns landing on the same number is stronger than either one alone. The annual monthly-equivalents assume the ten-months-billed-for-twelve structure reported by felloai.com on 2026-09-23, with the arithmetic verified independently here ($22 x 10 / 12 = $18.33; $99 x 10 / 12 = $82.50; $990 x 10 / 12 = $825.00). The vendor's own FAQ confirms annual billing exists, and the numeral did not render in a plain extract, so treat the ten-month structure as reported rather than vendor-confirmed.

The finding that no page in this result set publishes: the ladder stops getting cheaper at Pro. From Pro to Business the price rises tenfold, and the cost per finished minute does not move at all. Every tier from Pro upward costs $0.17 per minute, and the vendor's own extra-minute column independently agrees at ~$0.17 for all three. Above Pro you are not buying a cheaper minute. You are buying seats, voice slots, concurrent request headroom and audio format.

The crossover points, which are the actual decision

These three thresholds are where the article stops being a table and starts being advice. Each one answers the same question: at what monthly volume does the next tier become cheaper than topping up the current one?

Read the first row properly, because it is the one most teams will act on. Pro costs $77 more than Creator, and at Creator's own extra-minute rate that $77 buys about 428 minutes of top-up. Creator already includes 121, so the two paths cost the same at roughly 549 finished minutes a month. Below 549, Creator plus pay-as-you-go is cheaper. Above it, Pro is cheaper, and it also brings 44.1kHz PCM audio output via the API, 160 custom voice slots and 10 concurrent requests. That is the shape of every step on this ladder: the crossover is real, and so are the features bundled with it.

Three budgets, and the tier each one should actually buy

Volumes here are finished minutes, meaning output you keep, not generations you attempted.

A solo creator shipping 10 finished minutes a month. The tier usually sold is Creator at $22, and that is the wrong first purchase. Starter at $6 covers 30 minutes, three times the need. The uncomfortable number is the effective rate: at 10 minutes a month Starter costs you $0.60 per finished minute, three and a half times its steady-state rate, and Creator at that volume costs $2.20 per finished minute. At this size your floor is the subscription, not the narration. The one thing that overrides this is voice cloning: Professional Voice Cloning gates at Creator and does not exist below it, so a creator who needs their own trained voice is buying Creator for that feature, not for the minutes.

A five-person team shipping 60 finished minutes a month. Creator at $22 covers 121 minutes at $0.18 per minute, so on volume alone Creator is correct. The trap is seats, not credits. Every self-serve tier from Free through Pro carries one seat. Scale carries three. Business carries ten. There is no self-serve tier whose seat count matches a five-person team, so the team either shares a single login or moves to Business for seat management. A 60-minute requirement and a five-person requirement point at different tiers, and any page that answers this with one plan name is answering half the question.

An agency shipping 300 finished minutes a month. Creator's 121 minutes leaves a 179-minute gap. Two paths: Creator plus pay-as-you-go at 179 x ~$0.18, which is $32.22, for a monthly total of $54.22, or Pro at $99 flat. At 300 minutes the cheaper path is Creator plus top-up, and the reason to buy Pro anyway is the API audio format and the voice-slot count rather than the per-minute rate. That is the opposite of what the ranking explainers advise, and it follows directly from the table above.

Where the ladder beats the alternatives, and where it loses

Below roughly 30 finished minutes a month, the cheapest option is not a subscription at all. It is a one-off narration you commission per project, or a free-tier experiment you do not publish commercially. Once you cross into recurring, licensed output the subscription becomes the sane buy, and above 549 minutes a month the ladder itself is hard to beat on rate.

There are two honest exceptions. Audiobook-scale work is a rights and quality question before it is a rate question. A creator thread currently ranking at position 5 for this query puts human narration on ACX and Audible at $200 to $400 per finished hour and compares it against an ElevenLabs subscription. Treat that as a practitioner claim from a creator forum rather than a market rate, and treat the comparison as incomplete: a narrator is a licence, a performance and a delivery schedule in one purchase, and a synthetic voice is none of those until you assemble them. At long-form volume the two are not the same product.

The second exception is concurrency, and it is invisible on the pricing card. Voice agents and real-time applications fail on concurrent request limits before they fail on credits: 2 concurrent requests on Free, 5 on Creator, 10 on Pro, 15 on Scale, 25 on Business. If your build needs 20 simultaneous conversations, no amount of credit arithmetic gets you there below Business.

For the other meter inside your marketing budget, the same staleness problem is worth reading about in another AI tool whose ladder is explained third-party, where the field inherited a top tier the vendor stopped selling.

API access, seats and commercial rights

Three things sit under the plan ladder that the pricing card does not foreground, and each one has cost or compliance consequences.

Commercial rights begin at the first paid tier, and the terms say it plainly. The vendor's Terms of Use, non-EEA version, Last Updated 31 March 2026, section 3(c), splits users into Free Users who "may only use the Services for non-commercial purposes" and Paid Users who "may use the Services for commercial purposes". Starter at $6 is where that line falls, and the pricing page corroborates it: "Commercial License" is a Starter plan bullet and appears nowhere in the Free plan's list. If you plan to publish client work, the free tier is not a trial of the paid tiers. It is a different licence.

You own the output, and you grant a licence back. Section 4(c) of the same document retains all rights in your input and your output with you, and excludes the vendor's foundational models from that grant. Section 4(d) grants ElevenLabs a licence over your content described as perpetual, worldwide and sub-licensable. No page in the top fifteen mentions either clause. Read both before you clone a spokesperson's voice.

The API is a different meter, billed in dollars rather than credits. Vendor statement, read 2026-09-29: "API usage is billed in US dollars, not credits." The rates are per 1,000 characters: $0.10 on the multilingual models, $0.05 on Flash and v3 Conversational, which the vendor pairs with ~$0.10 and ~$0.05 per minute. Speech to Text runs $0.22 per hour on Scribe and $0.39 per hour on Scribe v2 Realtime. The Speech Engine for voice agents is $0.080 per minute included and $0.160 per minute burst. Music is $0.150 per minute. The important comparison is direction, not level: on the API path your model choice moves the bill by a factor of two, while the plan ladder above Pro moves the per-minute rate by nothing at all.

Three operational rules complete the picture. Credits roll over up to twice the monthly quota, capped at three times the quota total, which the vendor illustrates on Starter: 30,000 x 3 = 90,000 maximum. Rollover does not apply to the free tier, and cancelling or downgrading forfeits unused credits. Pay-as-you-go is stored as a dollar balance and converted at your current plan's rate, so it survives a plan change, with a 12-month life. And unused credits are forfeited if the account closes, per section 3 of the terms. If you are moving credits around between plans, plan the move rather than discovering it in a support ticket.

If your API usage runs through a platform rather than a direct integration, how metered API pricing works covers the billing mechanics that apply to any per-unit model, and seat-based plans covers the seat arithmetic this page just showed you at Scale and Business.

The bottom line: one recommendation per volume band

Under 30 finished minutes a month, buy the smallest tier that covers your need or commission narration per project, and check whether you actually require voice cloning before you spend more than $6. Between 30 and 549 minutes, Creator is the tier, and the volume is rarely the reason to leave it. Above 549 minutes the ladder hands you Pro, and above 1,776 it hands you Scale, but check the seat count first, because seats and minutes cross over at different points and a five-person team hits the seat wall long before it hits the minute wall.

The seat wall and the minute wall do not arrive at the same volume, which is the part of this ladder worth taking away: a five-person team can hit the seat limit well before it reaches the 549 minutes that justify Pro, so which tier is cheaper depends on which wall you hit first. Voice is one channel inside a campaign rather than a subscription with its own renewal date, which is why AI asset generation belongs in the same workspace as the narration it accompanies and why where a voice tool sits in the stack is a budget question rather than a tooling one. If you are buying per-seat tools for each asset type, the pricing model itself is the thing to compare, not the rate per minute. Allable starts free at 300 credits, a one-time allowance with no recurring grant, so the 1,000 credits a month only begin on the paid tiers: Pro at €31/month billed annually (€37 monthly) and Business at €91/month billed annually (€107 monthly).