Skip to content

ElevenLabs API Pricing: What a 30-Second Voiceover Costs

ElevenLabs API prices per model on 1 November 2026, how characters are billed, and a real 30-second ad script counted and priced, direct and via Aitachyon.

Get an API key, pay as you go from $10One key for video, image and voice models. Each call shows its price before it runs.
Get an API key
By Robin MFounder of Aitachyon

Published

Updated

Reading time6 min

Read inEnglishFrançais

ElevenLabs API pricing is per character of text you send, not per second of audio you get back. On the API pricing page, checked on 1 November 2026, Eleven v4 costs $0.08 per 1,000 characters and v4 Turbo $0.04. A 30-second ad read is around 500 characters, so one take on v4 costs about four cents.

Four cents is rarely the number that decides anything. Below: how the billing works (characters, credits, the two price tiers), a real 30-second script I wrote for this article, counted and priced take by take, and the same read through Aitachyon, the studio I run, which bills per block of characters and costs more per take. You should come out able to price a batch of reads to the cent and choose where to run it.

How ElevenLabs bills text to speech

The API pricing page says it in two lines: "Text to Speech is billed per character" and "API usage is billed in US dollars, not credits." The rate depends on the model, not on the plan: every plan column shows the same price per 1,000 characters, and there are two tiers. The model_id values and the caps per request come from ElevenLabs' models documentation.

Modelmodel_idPer 1,000 charactersMax per request
Eleven v4eleven_v4$0.0810,000
Eleven v4 Turboeleven_v4_turbo$0.04not listed
Eleven v3eleven_v3$0.085,000
Multilingual v2eleven_multilingual_v2$0.0810,000
Flash v2.5eleven_flash_v2_5$0.0440,000

On the API, the "model multiplier" is simply that factor of two: v4 Turbo, v3 Conversational and Flash cost half of v4, v3 and Multilingual v2. The models page also separates them by language coverage (90+ languages for v4, 70+ for v3, 32 for Flash v2.5, 29 for Multilingual v2) and lists the old Turbo v2.5 as deprecated in favour of Flash v2.5.

One default to override: the Create speech reference sets model_id to eleven_multilingual_v2 when you omit it. You pay the $0.08 tier and get the 29-language model, not v4. Set it on every call.

Credits are the app's unit, not the API's

ElevenLabs' main pricing page sells the same tiers in credits: Starter is $6 a month for 30,000 credits, Creator $22 for 121,000, Pro $99 for 600,000. There, text to speech costs 1 credit per character on Multilingual v2, and Flash and Turbo cost "between 0.5 and 1 credit per character" through the API. The help article Have characters changed? adds the other multiplier: some shared voices in the Voice Library carry a custom rate that raises credit consumption.

The API pricing page shows the same $6 Starter tier as 75,000 characters of the $0.08 models, which is exactly $6 at $0.08 per 1,000. Creator's $22 shows as 275,000 characters, Pro's $99 as 1,238,000. Read that way, the plan fee prepays usage at the list rate. For API work, budget from the API page. The commercial licence is listed on the Starter card and up, not on Free.

From seconds to characters

Scripts are planned in seconds and billed in characters, so you need a conversion with a source. ElevenLabs gives one: its API pricing table pairs 10,000 characters with "~10 min", and its help article on maximum characters per request says the same of v4's 10,000-character cap. That is about 1,000 characters a minute, or 500 for a 30-second read. It is an estimate: a slow, warm voice reads fewer characters in 30 seconds than a fast hook. Time the rendered file before you lock the edit.

Get an API key, pay as you go from $10One key for video, image and voice models. Each call shows its price before it runs.
Get an API key

A 30-second ad script, counted

Here is the script, for a fictional hand coffee grinder I made up for this example:

Most bad coffee starts at the grinder, not the beans. Meet Ostra, a hand grinder with forty steel burr settings, from espresso fine to French press coarse. Twist the dial, turn the crank, and you have fresh grounds in under a minute. No cord, no plug, no noise to wake the house. It fits in a backpack, comes apart without tools, and goes back together in seconds. Ostra. Grind it fresh, taste the difference. Order this week and your first bag of beans ships free, wherever you are.

Counted in JavaScript with text.length, every space and punctuation mark included: 483 characters, 88 words. At ElevenLabs' 1,000 characters a minute, about 29 seconds.

  • One take on v4: 483 x $0.08 / 1,000 = $0.0386.
  • One take on v4 Turbo: 483 x $0.04 / 1,000 = $0.0193.

Now a realistic job: four voices auditioned on v4 Turbo, then three final takes on v4, in English and in French. The French version I wrote for the French edition of this article is 510 characters, about 6% longer, which is why each language gets priced from its own text.

  • English: 4 x $0.0193 + 3 x $0.0386 = $0.1932.
  • French: 4 x $0.0204 + 3 x $0.0408 = $0.2040.
  • Total: $0.40 for 14 takes.

The voice is the cheapest line of an ad. The expensive habit is re-rendering a whole script to fix one sentence. The Create speech endpoint takes previous_text and next_text, so you can render one sentence with its neighbours as context, and a seed for best-effort repeatable output (ElevenLabs says determinism is not guaranteed).

Running a batch without 429 errors

A batch is limited by concurrent requests. ElevenLabs' 429 help page lists 2 on Free, 3 on Starter, 5 on Creator, 10 on Pro and 15 on Scale and Business, and warns the figures may change. It names two causes: too_many_concurrent_requests, which means your pool is too large for your plan, and system_busy, where a retry generally succeeds. Size the worker pool to the plan, back off on the second, and log characters and cost per file so the invoice can be reconciled.

The same voiceover through Aitachyon

Aitachyon bills ElevenLabs reads per started block of 450 characters: $0.108 a block on 1 November 2026, as shown on the ElevenLabs voiceover price page. The derivation is the public rule from the MCP page, provider cost times 3 on audio: 450 x $0.08 / 1,000 = $0.036, times 3. Reads run on eleven_v4, falling back to eleven_v4_turbo if a v4 call fails.

The 483-character script is two blocks, so $0.216 a take: 5.6 times the direct v4 price. Blocks reward editing. Cut "no plug", "and goes back together in seconds" and "wherever you are", and the script drops to 424 characters, one block, $0.108. The French version trims the same way to 443.

The flow, documented on the API reference: GET /api/voiceover/voices?provider=elevenlabs lists the voices and their voiceId (free, with your key); POST /api/voiceover/sample plays a fixed sentence in that voice at no charge, which replaces the paid audition round; POST /api/voiceover/generate with provider: "elevenlabs", the voiceId, the text and a projectId renders the take. Over MCP, the same steps are the voices_list and voiceover_generate tools.

Priced on the job above: six final takes cost $1.30 at full length, or $0.65 with both scripts trimmed to one block, against $0.24 for the same six takes direct on v4. The same endpoint also offers Google voices at $0.0405 a block and Fish Audio at $0.021 to $0.0225.

My honest read: if voice is all you need, call ElevenLabs directly; it is several times cheaper. Going through Aitachyon makes sense when the voiceover is one step of a job that also renders video and images on the same balance, with failed jobs refunded and the price of each call known before it runs. The video side of that arithmetic is in AI video generation API pricing, lab by lab, and top-ups are explained on the pricing page.

Robin MFounder of Aitachyon
How the studio works

One key for every model.

Video, image and voice models behind one API key and one MCP server. Pay as you go from $10: each call shows its price before it runs, and a failed job is refunded to the cent.

Get an API key

How the API and the MCP work