Seedance 2.5 from ByteDance is officially available on Apiframe.

ElevenLabs API Guide: Pricing, Voice Cloning & Code (2026)

ElevenLabs API explained: voice cloning, pricing, access, and code for 2026.

Apiframe Team Last updated November 11, 2026 August 9, 2026 · 6 min read Beginner
ElevenLabs API Guide: Pricing, Voice Cloning & Code (2026)

ElevenLabs has become one of the go-to platforms for AI voice: text to speech that sounds genuinely human, voice cloning that takes seconds to set up, and a full toolkit for building voice agents. Here is what the API offers, what it costs, and how to make your first request, with a working code example. We will also cover why voice pairs so naturally with AI video and avatar work, since that combination is where a lot of interesting products are being built.

What the ElevenLabs API Is

The API gives you the same voice engine behind the ElevenLabs web app. That covers text to speech, voice changing (feed in one recording, get it back in a different voice), voice cloning, voice design, and a layer for building agents that hold real phone or chat conversations.

It is built for developers adding narration, dubbing, voice assistants, or voiceovers to their own products rather than using ElevenLabs as a standalone tool. You authenticate with a key, send your text and a chosen voice, and get audio back. From there you can build anything from an audiobook workflow to a support voice bot.

If voice is only one part of a product that also handles images or video, a unified AI API makes the rest easier to manage in one place. Our explainer on what a unified AI media API actually is covers the basics if the term is new to you.

Key Features

Models and languages

ElevenLabs runs several text-to-speech models at once, and picking the wrong one is the most common mistake. Language coverage and the amount of text each will accept in one request vary widely, so check both before committing.

Eleven v4 is the current flagship and the most expressive of the group. It covers 90 or more languages and takes up to 10,000 characters per request, roughly ten minutes of audio. If you are starting fresh, start here.

Eleven v4 Turbo is the real-time version of v4, at around 100 milliseconds, with the same 90 plus language coverage. This is now the model ElevenLabs points to for voice agents.

Eleven v3 is still widely used and takes up to 5,000 characters, covering 70 or more languages, but ElevenLabs now labels it previous generation. There is a conversational variant at around 165 milliseconds.

Multilingual v2 takes up to 10,000 characters and remains the steadiest on long narration, but it is also previous generation and covers only 29 languages. It does handle numbers, dates, and currencies more reliably than the Flash models, which matters if your text contains phone numbers or prices.

Flash v2.5 trades expressiveness for speed and price, running at around 100 milliseconds and accepting up to 40,000 characters, by far the largest window. It costs half as much per character through the API. The catch is that it covers 32 languages, and it does not tidy up numbers by default, so a phone number may be read as one long figure.

You will still see Turbo v2.5 recommended in older write-ups. ElevenLabs has deprecated it and says the Flash models do the same job with lower latency, so use Flash instead.

The voice library is large, with thousands of community voices you can use directly alongside any you create yourself. If you are weighing ElevenLabs against other providers, our roundup of the best AI media generation APIs and our guide on how to choose an AI media API both help frame the decision.

Voice cloning and voice design

This is what most people come to ElevenLabs for. Instant Voice Cloning builds a usable clone from a short sample and starts on the Starter plan. Professional Voice Cloning gives a closer match and unlocks on Creator and above. Voice Design generates a brand new voice from a written description, useful when you want something distinctive without recording anyone. If you are cloning a real person's voice, get their consent in writing, and if clones will be used in anything customer-facing, our guide to AI content moderation covers the checks worth adding.

Real-time and streaming

For voice agents and live applications, the delay between request and first audio matters more than anything else. ElevenLabs offers streaming endpoints and a persistent connection option, with v4 Turbo and Flash v2.5 both tuned for real time. This is what powers the agents product, which handles phone calling through Twilio or a standard phone connection and can be pointed at your own documentation so it answers questions accurately.

If speed is central to what you are building, our guide to latency across media APIs covers how to plan around it.

Pricing

ElevenLabs runs on a monthly credit allowance, and how far it stretches depends on the model. On v3 and Multilingual v2, one character costs one credit. Flash runs at 0.5 to 1 credit per character through the API, so the same balance can go up to twice as far. One thing to watch: every product draws from the same pool, so transcription, music, and sound effects all eat into the credits you were planning to spend on speech.

Here is the tier breakdown:

TierPrice/monthCredits included
Free$010,000
Starter$630,000
Creator$22121,000
Pro$99600,000
Scale$2991,800,000
Business$9906,000,000
EnterpriseCustomCustom

How many requests you can run at the same time is set by your plan, and for agent workloads that ceiling usually bites before your credit balance does. On Multilingual v2 it runs 2 on Free, 3 on Starter, 5 on Creator, 10 on Pro, and 15 on both Scale and Business. The Flash models get double those numbers. Worth knowing that this is not the same as the number of conversations you can hold at once: ElevenLabs reckons a limit of 5 supports roughly 100 simultaneous conversations, because each one only generates audio in short bursts.

The free tier does not include commercial usage rights, so anything you plan to ship needs at least Starter. Past your monthly allowance, pay-as-you-go runs roughly $0.05 per 1,000 characters on Flash and $0.10 per 1,000 on Multilingual v2 and v3. Annual billing works out to about two months free. On paid plans unused credits roll over for up to two months, so a balance can reach three times your monthly quota, but cancelling or downgrading forfeits them, and the free plan gets no rollover at all.

One budgeting detail worth knowing: regenerating audio costs the same as generating it the first time. A mispronounced name or a take that lands wrong still burns credits, so build slack into your estimates, especially if you are generating narration that needs to be exactly right. For the wider licensing picture, see our guide on copyright and commercial rights for AI-generated content.

Getting Access

  1. Create a free account at elevenlabs.io. No card is needed for the free tier.
  2. Go to your account settings and generate a key. You can limit a key to specific permissions and set credit caps or allowed IP addresses on it, which is worth doing before it goes anywhere near production.
  3. Send the header xi-api-key on every request.
  4. Pick a voice ID from the library (or one you have cloned), pick a model, and you are ready to call.

Your First Call

Here is a minimal example using the official Python library to turn text into speech and save it as an MP3.

python
import os
from dotenv import load_dotenv
from elevenlabs.client import ElevenLabs

load_dotenv()

elevenlabs = ElevenLabs(api_key=os.getenv("ELEVENLABS_API_KEY"))

audio = elevenlabs.text_to_speech.convert(
    text="The first move is what sets everything in motion.",
    voice_id="JBFqnCBsd6RMkjVDRZzb",  # "George" from the voice library
    model_id="eleven_v4",  # swap for eleven_flash_v2_5 to halve the cost
    output_format="mp3_44100_128",
)

with open("speech.mp3", "wb") as f:
    for chunk in audio:
        f.write(chunk)

Install the library first with pip install elevenlabs python-dotenv. If you would rather skip it, the same request works as a plain cURL call to https://api.elevenlabs.io/v1/text-to-speech/{voice_id} with your key in the xi-api-key header and a JSON body containing text and model_id.

The model is the one line worth revisiting. Swapping eleven_v4 for eleven_flash_v2_5 halves what you pay per character and cuts the wait, at the cost of some expressiveness and a lot of language coverage. For bulk narration where the delivery does not need to carry emotion, that trade is usually worth making. Test both on your actual script before deciding, since voice previews rarely sound like your own text.

For long jobs where you would rather not hold a connection open, our guide on having results sent to your server automatically covers the alternative pattern.

Alternatives

OpenAI's text-to-speech models are the usual comparison. They are noticeably cheaper per character and simple to use, but they do not offer voice cloning and generally sound flatter over longer passages. Google Cloud TTS, Amazon Polly, and newer arrivals like Cartesia tend to be faster and cheaper still, again without the cloning quality that is ElevenLabs' main draw. Our roundup of the best AI media generation APIs covers the wider field. In short, if cloning or emotional delivery is the point, ElevenLabs is worth the premium. If you need plain, clear narration at volume, you are paying for range you will not use.

If your project needs music alongside voice, whether background beds, jingles, or full songs, see our guides to the Suno API, Udio API, and Mureka API, our complete AI music API guide, or the roundup of the best AI music generation APIs. ElevenLabs has its own music model too, billed from the same credit pool.

Voice is rarely the whole product. Teams building avatars, dubbed video, or talking-head content usually pair a voice API with an image or video model for the visuals and lip movement. Our AI avatar API guide covers that pairing, and the AI video API guide covers the video side. Worth noting that several video models now generate their own synced audio, Kling 3.0 among them, so check whether you need a separate voice step at all before building one.

To be clear about scope: ElevenLabs is not available through Apiframe, so voice stays a direct integration. What a unified API handles is everything around it, the image, video, and music models, behind one key instead of one account per vendor. If voice is one piece of something larger, that still removes most of the integration work, just not this piece of it.

FAQ

Is there a free ElevenLabs API tier?

Yes. The free plan includes 10,000 credits a month, roughly 10 minutes of speech on the standard models, but it carries no commercial usage rights, so it suits testing and personal projects only. Free credits do not roll over. For how free allowances compare across the wider media API field, see our guide to what is actually free in generation APIs.

How does pricing work per character?

Credit based. On v3 and Multilingual v2, one credit covers one character. Flash uses 0.5 to 1 credit per character, making it the cheaper option when you do not need maximum expressiveness. Past your monthly allowance, overage runs about $0.05 per 1,000 characters on Flash and $0.10 on the standard models.

Which model should I use?

For produced audio such as audiobooks, ads, and narration, use v4. For voice agents and anything live, use v4 Turbo. For bulk work where cost matters more than delivery, use Flash v2.5, but check your languages are among its 32 first. If your text is full of phone numbers, dates, or prices, Multilingual v2 reads them more reliably than Flash does.

Can I clone a voice using the API alone?

Yes. Instant Voice Cloning starts on the Starter plan and works from a short sample. Professional Voice Cloning, which gives a closer match, needs Creator or above.

Does it support real-time or streaming audio?

Yes. There are streaming endpoints and a persistent connection option, with v4 Turbo and Flash v2.5 built for low delay. Keep your plan's simultaneous request limit in mind, since that is usually what caps how many live conversations you can run.

How many requests can I run at once?

On Multilingual v2 it is 2 on Free, 3 on Starter, 5 on Creator, 10 on Pro, and 15 on Scale and Business. Flash doubles each of those. Going over the limit queues the request rather than failing it, typically adding about 50 milliseconds. Building with more than just voice? Our AI video generation API guide walks through the top video models and how to integrate them. For images, the AI image API guide covers everything from Midjourney to Flux. And for one key across image, video, and music, the unified AI API guide is the place to start.

You can also get an API key and start with 100 free credits if you need video, image, and music models in one integration alongside your ElevenLabs setup.

Apiframe Team

The team behind Apiframe - making AI generation accessible to everyone.

Power your next AI product with Apiframe.

Instant access to 70+ media models through a single API. Start free and scale when you're ready.