ElevenLabs on OpenRouter is now live, giving developers direct access to nine text-to-speech models and two speech-to-text models through a single API endpoint. Announced on the OpenRouter blog, the addition means teams running language models through the routing platform can now generate spoken audio and transcribe conversations without maintaining a separate ElevenLabs subscription.
All ElevenLabs audio models on the platform are discounted by 50 percent off OpenRouter list prices through October 19 at 8:00 AM PT. The launch brings production-grade speech synthesis and transcription into standard developer workflows using existing OpenRouter account balances and authentication keys.
What Changed: ElevenLabs on OpenRouter Models and Pricing
Developers access the new audio tools by making requests to two primary endpoints: POST /api/v1/audio/speech for synthesis and POST /api/v1/audio/transcriptions for transcription. By integrating ElevenLabs on OpenRouter, teams avoid managing separate API keys, multiple credit pools, and distinct vendor agreements.
The text-to-speech collection spans nine distinct models divided across two pricing tiers. Eleven v4, Eleven v3, and Eleven Multilingual v2 bill at the full per-character rate. Eleven v4 Turbo, Eleven v3 Conversational, Eleven Flash v2.5, Eleven Flash v2, Eleven Turbo v2.5, and Eleven Turbo v2 bill at half that rate.
For audio input, OpenRouter added two speech-to-text models: Scribe v2 and Scribe v2 Medical. Both models process audio using the same request shape and identical per-second pricing. The company notes that Scribe v2 Medical reduces transcription errors on clinical vocabulary by 35 percent compared to the base Scribe v2 model. Realtime WebSocket streaming for Scribe v2 is not included in this release.
- Eleven v4 handles up to 10,000 characters per request for expressive narration and supports delivery tags like [whispering] and [laughing].
- Eleven v4 Turbo processes up to 10,000 characters per request, engineered specifically for low-latency voice agent responses at half the base rate.
- Eleven Multilingual v2 offers 10,000 characters per request with cross-language speed control.
- Eleven Flash v2.5 supports up to 40,000 characters per request across 32 languages for high-volume, cost-effective generation.
- Scribe v2 provides speech transcription across more than 90 languages with word-level timestamps and speaker diarization.
Why ElevenLabs on OpenRouter Matters for Builders
Before this release, building an interactive voice agent required separate integrations with a model gateway, a speech provider, and an audio transcription service. With ElevenLabs on OpenRouter, developers can run an entire voice pipeline through one provider interface. A single API key now authenticates prompt completion, user transcription, and spoken audio delivery.
The integration also preserves specialized provider parameters. Developers can pass options like previous_text and next_text to ElevenLabs models using the provider.options.elevenlabs parameter block. This ensures consistent voice pitch and intonation when stitching together segmented paragraphs for longer narrations.
Similarly, Scribe v2 supports speaker diarization through the provider configuration block. Setting diarize to true and passing num_speakers allows applications to break meeting recordings into distinct speaker turns, each tagged with word-level timing and audio events such as laughter.
Practical Steps to Use ElevenLabs on OpenRouter
Calling ElevenLabs on OpenRouter requires only your standard authorization bearer token. To synthesize an audio file, developers send a JSON payload specifying an ElevenLabs model ID, input text, output format, and one of 21 premade voice names, such as george.
For long narrations, scripts should be split at paragraph breaks because Eleven v4 limits individual requests to 10,000 characters while Eleven v3 caps requests at 5,000 characters. Passing surrounding text chunks in the options block maintains continuity between generated files.
Building a voice agent involves chaining three calls: transcribing incoming user audio with Scribe v2, sending the text to any chat model in the OpenRouter catalog, and feeding the reply into Eleven v4 Turbo. Because responses bill against your existing account balance, managing agent expenses requires monitoring only a single usage dashboard.
File uploads for transcription currently support up to 25 megabytes via multipart form data, which accommodates roughly 27 minutes of 128 kbps MP3 audio. Longer recordings require passing a publicly accessible audio URL rather than a local file upload.
FAQs
Do I need an ElevenLabs account to use ElevenLabs on OpenRouter?
No, you do not need an ElevenLabs subscription or account. Calls to ElevenLabs on OpenRouter authenticate using your existing OpenRouter API key and deduct from your standard platform credits.
Which ElevenLabs voice model is best for conversational agents?
Eleven v4 Turbo is designed specifically for low-latency voice interactions. It bills at half the per-character cost of Eleven v4 and delivers quick audio responses suitable for live agents.
Does ElevenLabs on OpenRouter support real-time transcription over WebSockets?
No, real-time WebSocket transcription for Scribe v2 is not included in this launch. Transcription requests currently operate over standard HTTP POST calls with uploaded files or public audio URLs.
What This Changes for Client Builds
In client projects, simplifying API architecture directly reduces system failure points. Wasif builds scalable workflows using AI automation and Agentic AI Systems & AI Workforce architectures, where managing separate billing accounts and credential rotations adds unnecessary operational overhead.
Consolidating speech synthesis, transcription, and language models under one router makes automated voice agents faster to deploy and easier to maintain. Because developers can freely swap the underlying chat model without altering audio endpoints, workflows gain flexibility as underlying models improve.
The temporary 50 percent launch discount makes this an ideal window to benchmark ElevenLabs on OpenRouter across automated voice agents, customer support workflows, and content generation pipelines. If you want to integrate voice capabilities into your business systems, reach out to Wasif Ahmed to discuss your architecture.
