ElevenLabs Pricing: The Same Plan Buys 1x to 16x More Through the API

Last updated: August 12, 2026

ElevenLabs pricing covers six self-serve plans, from $0 to $990 per month, all metered in credits. The website and the API price the same models at different rates, and ElevenLabs says so in its Pay As You Go documentation. What no page states is the size of that difference. Working it out from the two published tables gives a spread from no difference at all on sound effects to roughly sixteen times on transcription.

Verdict in one sentence: pick the surface your production will actually run on before you pick the plan, because the same subscription is worth a very different amount depending on the answer.

Two capacity tables, one plan

The ElevenCreative pricing page gives each plan a monthly credit quota and lists what each product costs in credits. Creator gets 121,000 credits a month, text to speech costs one credit per character on the standard multilingual models, and transcription costs 330 credits per minute.

The ElevenAPI page describes the same six plans, at the same six prices, using a different unit. It prices each model in dollars, then shows how much of that model each plan includes. For Creator it shows 220,000 characters of Multilingual text to speech and 100 hours of Scribe v2 transcription.

Neither page is wrong. They are answering the same question through different meters, and ElevenLabs states in its Pay As You Go documentation that API usage carries lower rates than UI usage. The rates on both pages are published. The relationship between them is not.

Sources: ElevenLabs, ElevenCreative pricing and credit FAQ, accessed 12 August 2026 · ElevenLabs, ElevenAPI model pricing, accessed 12 August 2026 · ElevenLabs, Pay As You Go documentation, accessed 12 August 2026

The spread, product by product

Both tables answer the same hypothetical: spend the entire plan on one product, and how much do you get. Applying that assumption on both sides makes the comparison clean, because the assumption cancels out.

Here is the Creator plan at $22, across every product where both pages publish a figure.

Website column is our arithmetic: 121,000 credits divided by the credit cost published for each product. API column is read directly from the ElevenAPI table. Ratios are ours. Flash and Turbo are excluded because their discount is stated to apply to API usage only, which makes the two sides non-comparable.

Sound effects is the control case. The two pages produce the identical figure, 605 generations, which shows the difference is not an artifact of the method.

From there the gap widens unevenly. Music and dubbing move by about nine percent. Voice changer moves by half. Text to speech moves by 1.8 times. Transcription moves by more than sixteen.

A team that budgets transcription from the credit table and then builds on the API will find its plan stretches far further than expected. A team that does the reverse will run out.

Sources: ElevenLabs, ElevenCreative credit costs, accessed 12 August 2026 · ElevenLabs, ElevenAPI included quantities by plan, accessed 12 August 2026

Why every API figure equals the plan price

The API table is not a second allowance sitting on top of the credit quota. Multiply any included quantity by its own published rate and the answer is the plan price.

Arithmetic ours, from quantities and rates published on the ElevenAPI page. The pattern holds on the other paid tiers: Starter $6, Pro $99, Scale $299, Business $990.

This is the strongest reading of what the API page is doing, and it is worth stating plainly because it defeats a tempting misreading. The page is showing what your monthly fee is worth if you direct all of it at one product through the API. It is not promising that a Creator account holds 121,000 credits and 220,000 characters and 100 transcription hours as separate balances.

The finding survives that reading. If the API column is plan value at API rates, and the credit column is plan value at website rates, then the ratio between them is the price difference between the two channels. That is precisely the number a buyer needs and precisely the number neither page prints.

Source: ElevenLabs, ElevenAPI model pricing table, accessed 12 August 2026

What the spread proves and what it does not

It proves a difference in published capacity between two official surfaces for the same subscription. Both figures are current, both are the vendor’s own, and the assumption behind them is identical.

It does not prove that a bill will come in sixteen times higher. Nothing here was measured inside an account. The published tables describe capacity, and an invoice describes charges, and this briefing did not join the two.

The API does expose some per-request billing metadata. ElevenLabs documents a character-cost response header alongside request-id and x-trace-id, described as generation metadata including character costs. That header is a character count. The documentation does not define it as a dollar amount or as a credit debit, and the workspace usage endpoint reports credits grouped into time buckets rather than per request. So the pieces exist at different levels of aggregation, and no documented key joins a single request to a credit movement or an invoice line.

Closing that gap needs a paid account, a controlled run, and an invoice. That is a different piece of work, and until it is done, the honest position is that the published tables disagree by a knowable amount and the billing behavior behind them is untested.

Sources: ElevenLabs, API reference introduction, Tracking generation costs, accessed 12 August 2026 · ElevenLabs, Usage analytics documentation, accessed 12 August 2026

Agents runs on a third meter

The ElevenAgents page adds a complication that neither of the other two carries. Creator there shows 275 included call minutes, additional minutes at $0.08, and burst pricing at $0.160, which is double.

Those base call minutes sit outside the shared credit pool. The page separates them from credits, prices the LLM component as a deduction from ElevenLabs credits, and treats telephony as billed separately at cost.

The result is that an agent deployment carries three cost layers rather than one: hosting minutes on their own meter, model usage against the credit pool, and carrier charges outside both. A team budgeting agents from the minute count alone will be short.

Source: ElevenLabs, ElevenAgents pricing and cost calculator, accessed 12 August 2026

What happens to a balance you do not spend

Three balances follow three different rules, and mixing them up is expensive.

Sources: ElevenLabs, Billing documentation, accessed 12 August 2026 · ElevenLabs, Pay As You Go documentation, accessed 12 August 2026

The prepaid row carries the most risk. Section 6(b) of the Terms describes prepaid credits as a limited, revocable, prepaid right to access services, and states they are not money, not a deposit, and not stored value. They expire twelve months after purchase, with no obligation on ElevenLabs to warn you first. Because the balance is held as a dollar figure and converted at the current plan’s rate, a downgrade lowers the credit count on screen while the dollar value stays put.

Refund language differs by document and by region rather than contradicting itself outright. The billing documentation describes a fourteen day window where no quota was used. The non-EEA Terms say all sales are final with refunds at ElevenLabs’ discretion. The EEA Terms grant a statutory fourteen day withdrawal right with a model form attached. Which applies depends on where you are.

Sources: ElevenLabs, Terms of Service (non-EEA), 31 March 2026, Sections 3, 6(b), 6(e) · ElevenLabs, Terms of Service (EEA), 31 March 2026, Sections 6(b), 7

What you can and cannot decide from this

Four questions close the right column, and one paid Creator month answers all of them: record the starting balances, send fixed input through the API and through the web app, export the request log and the usage CSV, then compare both against the invoice.

Source: scope statement of this briefing. No ElevenLabs account was used.

A note for buyers in Japan

Japan falls outside the EEA, Switzerland, and the UK, so Japanese buyers are on the non-EEA Terms. Section 6(g) states that where the services are subject to Japanese consumption tax under the reverse charge mechanism, the customer self-assesses, reports, and remits the tax directly, and that ElevenLabs will not charge or remit JCT for those services. Whether that applies to a given purchase depends on the buyer’s own circumstances. Confirm it against an actual invoice with your tax adviser before treating the position as settled.

Source: ElevenLabs, Terms of Service (non-EEA), 31 March 2026, Sections 6(d) and 6(g)

METHODOLOGY

This is a document-first briefing. Future Stack Reviews holds no ElevenLabs account, ran no generations, and inspected no invoice, checkout, or usage ledger.

Every figure comes from a page opened on 12 August 2026: the three pricing tabs, the billing and Pay As You Go documentation, the usage analytics page, the API reference introduction, and the Terms of Service for both the EEA and non-EEA regions dated 31 March 2026. Links sit under the section each supports.

The website capacity column is arithmetic, dividing the plan quota by each product’s published credit cost. The ratios are ours. Both columns assume the whole plan is spent on one product, which is why they compare cleanly.

Absence statements are bounded to the pages listed above. Pricing on this vendor moved during 2026, so recheck before purchase. Our full evidence and tier rules are set out in the Future Stack Reviews methodology.

VERDICT

Choose the production surface before the plan. On sound effects the choice is free. On transcription it changes what $22 buys by more than an order of magnitude, and neither pricing page tells you that.

For a single-surface team, budget from the table that matches the surface and start with a capped first cycle. For a mixed workflow, or before an annual commitment or a meaningful top-up, run one paid month and reconcile it against an invoice. The capacity tables are good enough to size a plan and not good enough to close a finance review.