Runway's Eleven v4 speech API: request shape and October pricing

Runway added Eleven v4 to its developer API on September 30. Its release note identifies the model as eleven_v4, caps a script at 2,500 characters and says the credit rate will rise on October 12, 2026 Pacific Time. For developers adding speech generation to a production workflow, the useful question is what a valid request needs and how much the documented rate means in dollars. This guide is based on Runway's documentation checked October 6; BIG CHANGE did not make a paid API call or assess the audio.

The big change

• What changed: Runway now lists Eleven v4 as a text-to-speech model behind its existing task API. A script can include documented audio tags such as [laughs], [whispers] and [pause], within the 2,500-character limit.
• Why it matters: A team with a Runway developer project can request speech using a preset voice and retrieve the completed task's audio URL. Its integration and budget planning can use one documented request shape and a dated price schedule.
• What to watch: Runway says the rate changes from 2.2 to 5 credits per 1,000 characters on October 12, 2026 PT. Anyone estimating repeated generations needs to use the rate effective when the request is billed and check the live price before committing a budget.

Set up a project and key

Runway's setup guide says to create an account and project in the developer portal, then create a project API key under API Keys. Creating or disabling a key requires an Admin or Developer role. The key appears in plaintext once, so store it securely. For SDK use, Runway says to set RUNWAYML_API_SECRET; its SDK reads that variable automatically. Keep the secret out of source code.

The project also needs credits before it can use the API. Runway says the minimum initial payment is $10, at $0.01 per credit, through the project's billing tab. An Admin or Billing admin role is needed to add credits. That is a minimum credit purchase, not the charge for one speech request. The setup page does not describe a free first call.

Send a text-to-speech request

The Runway Python SDK's current request type requires model, prompt_text and a voice object for Eleven v4. On the wire, the corresponding JSON fields are model, promptText and voice, with voice.type set to runway-preset and a presetId. Bernard is one listed preset. The SDK service posts this request to /v1/text_to_speech and returns a new task response. Runway's API guide documents the host, bearer key and X-Runway-Version: 2024-11-06 header used for direct HTTP calls.

This is a documentation-derived request body, not a call we ran:

{
  "model": "eleven_v4",
  "promptText": "The recording begins now. [pause] Welcome.",
  "voice": {
    "type": "runway-preset",
    "presetId": "Bernard"
  }
}

Runway's September 30 changelog entry names [laughs], [whispers] and [pause] as supported examples and says Eleven v4 uses the same Runway preset voices as Eleven v3. Tags are part of the script; leave room for them within the stated 2,500-character maximum. The SDK also lists optional voice settings, language and neighboring text fields. Its schema does not establish that any setting guarantees a particular delivery, and this article makes no claim about voice quality or latency.

The initial response is a task, rather than audio bytes. Runway's output guide says to query GET /v1/tasks/:id; a succeeded task contains an output array of result URLs. Those URLs expire within 24–48 hours of accessing the API, so download the result into your own storage rather than using the temporary URL in a product. The general output example is a video task; it establishes the task response shape, while the models list identifies Eleven v4's output as audio. We have not observed an Eleven v4 response or verified its audio file format.

Calculate the dated price

Runway's pricing page and release note list 2.2 credits per 1,000 characters until October 12, 2026 PT, then 5 credits per 1,000 characters, with a one-credit minimum. At the published $0.01 per credit, those rates translate to $0.022 and $0.05 per 1,000 characters before any applicable tax. The later listed rate is about 2.27 times the earlier one.

For a transparent planning example, ten separate scripts of exactly 1,000 billable characters each would total 10,000 characters. Multiplying the published rates gives 22 credits, or $0.22, at the earlier rate and 50 credits, or $0.50, at the later rate. Each script is below the 2,500-character limit, and each calculated request is above the one-credit minimum. This is arithmetic from Runway's listed rates, not an observed invoice. The cited pages do not specify how billable characters treat tags or whether fractional per-request charges are rounded, so a production estimator should check actual usage records and current terms.

Sources & further reading

Runway's September 30 API changelog establishes the release date, model ID, tag examples, script cap and dated pricing statement. The setup guide explains project, key and credit prerequisites, including role permissions and the initial $10 payment. The pricing page supplies the credit-to-dollar conversion and listed Eleven v4 rates.

The Python SDK request schema and service implementation establish the field names, preset voice shape, endpoint path and task creation response inspected for this guide. Runway's getting started guide, model list and output guide supply the HTTP version header, audio output classification and task retrieval behavior. These are provider documents, not an independent performance test. Checked October 6, 2026.