Eleven v4 vs Eleven v4 Turbo: Speech Controls, Price and API

Eleven v4 and Eleven v4 Turbo generate speech from text with voice selection, audio tags, stability and similarity controls. On Raywake they take the same inputs, including a default Rachel voice and MP3 output. Raywake lists up to $0.13 for Eleven v4 and up to $0.07 for Turbo per run labeled 1,000 ASCII characters. Raywake does not list a maximum script length; try a short passage first.

Compare Eleven v4 and Eleven v4 Turbo when you want the same documented controls at different prices. Prices and settings are as of 31 October 2026. Compare both on your own script to judge speed and voice quality.

Eleven v4 and Turbo at a glance

How much do Eleven v4 and Turbo cost?

The available prices are upper bounds attached to a character-based label.

Do not describe these as fixed prices for every script. Quote the exact text and settings you plan to send.

Turbo has the lower listed amount for the same label. That can help prioritize a comparison, but a budget decision should still use matching requests. A different script, voice or output setting makes it harder to tell what caused a result or quote to differ.

Which one should you use?

Start with Turbo when the listed controls meet your needs and the lower listed price is useful. Compare Eleven v4 on the same script if you want to judge its delivery against Turbo. Keep the voice, normalization and output format constant during that comparison.

Listen for the things your project actually requires: a correctly pronounced name, a clear sentence ending or a suitable pause before a key point. A general “naturalness” impression is less actionable than a specific detail you can revise.

Both models support audio tags and IPA pronunciation. Check the supported syntax before adding tags or pronunciation instructions. Start with plain text, as in the runnable example, to hear the default delivery.

What do stability and similarity settings expose?

Both models list stability from 0 to 1, with 0.5 as the default. Both list similarity_boost from 0 to 1, with 0.75 as the default. Use these settings to compare how the voice delivers your script.

Begin with the defaults and change one value at a time. Keep the script identical when comparing settings, otherwise a new sentence rhythm can obscure the effect you are trying to judge.

The default voice is Rachel. The code uses that voice as a starting point. The optional language_code field accepts a string; check which language values Raywake supports before setting it.

How do normalization and audio formats work?

apply_text_normalization accepts auto, on or off and defaults to auto. Make the choice explicit if your script contains numbers, abbreviations or other text whose spoken form matters. Listen to those passages before assuming the intended reading was produced.

Both output-format lists include MP3, PCM, µ-law, A-law and Opus choices. The default is mp3_44100_128. Select an exact enum value supported by your playback or editing environment, rather than a generic format name copied from another model.

timestamps is a boolean that defaults to false. If you request timestamps, inspect their structure and alignment with the audio before building a parser.

How to call Eleven v4 through the API

The example quotes Turbo with an explicit voice and the default stability, similarity and format settings.

curl --fail-with-body https://api.raywake.com/v1/quotes \
  -H "Authorization: Bearer $RAYWAKE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"elevenlabs/tts/eleven-v4-turbo","input":{"text":"Welcome to the exhibition. Take a moment to look at the light on the water.","voice":"Rachel","stability":0.5,"similarity_boost":0.75,"apply_text_normalization":"auto","output_format":"mp3_44100_128"}}'

For Eleven v4, change the outer model ID to elevenlabs/tts/eleven-v4. Both operations accept the same input fields, which makes a controlled comparison straightforward.

The spoken script belongs in text. Read the quotes reference for the quote response and subsequent request flow. The sample does not submit a generation.

Common questions

Is Turbo proven to be faster?

The name alone does not establish a speed advantage. Compare completion times for matching requests to judge which suits your workflow.

Do both versions have the same controls?

Yes. Both have the same inputs, including voice, stability, similarity, normalization, formats, seed and timestamps.

Is the price label a maximum text length?

No. “1,000 ASCII characters” appears in pricing. Raywake does not list a maximum length for text.

Can I switch models without rewriting the payload?

For the fields shown here, yes. Replace the model ID and quote the request again, since the listed prices differ.