October 7, 2026
ElevenLabs voice generation costs less on OpenRouter: all models are 50% off
ElevenLabs text-to-speech and speech recognition cost 50% less on OpenRouter until Oct 19.
OpenRouter
@openrouter
ElevenLabs is officially available on OpenRouter. You can now call 9 speech synthesis models supporting up to 90+ languages and 2 speech recognition models with speaker labels and word-level timestamps. All @ElevenLabs models are available at a 50% discount only on OpenRouter until Oct 19, 08:00 US Pacific Time.
· 29.5K views
OpenRouter has added 9 ElevenLabs text-to-speech models and 2 speech recognition models. The higher-end text-to-speech models support 90+ languages.
OpenRouter already supported audio transcription through its API. ElevenLabs models are now available through it too: applications can turn Kazakh text into speech and transcribe recorded conversations with speaker labels and word-level timestamps.
Text-to-speech through the API. To generate audio, send a POST request to `/api/v1/audio/speech` with the fields `model`, `input` and `voice`. The parameter `response_format="mp3"` returns an audio stream that you can save to a file.
For Eleven v4, specify `model="elevenlabs/eleven-v4"`; for v4 Turbo, use `elevenlabs/eleven-v4-turbo`. Select a voice in the `voice` field; the catalog includes george, sarah and adam.
As of Oct 7, the OpenRouter catalog includes: - Eleven v4: 90+ languages, including Kazakh, up to 10,000 characters per request. - Eleven v4 Turbo: 90+ languages, including Kazakh. - Eleven v3. - Eleven v3 Conversational. - Eleven Multilingual v2: 29 languages, up to 10,000 characters per request. - Eleven Flash v2. - Eleven Flash v2.5: 32 languages, up to 40,000 characters per request. - Eleven Turbo v2. - Eleven Turbo v2.5.
ElevenLabs recommends Flash over Turbo for all use cases. Turbo v2 and v2.5 perform the same tasks as their corresponding Flash models, but Flash starts responding faster on average.
Transcribing a recording. Speech recognition is available through `elevenlabs/scribe-v2` and `elevenlabs/scribe-v2-medical`. Both models return word-level timestamps, can separate speakers on request and label sound events.
Send a POST request to `/api/v1/audio/transcriptions` with an OpenRouter Bearer key. Specify the model in `model`, pass base64-encoded audio in `input_audio.data` and specify the file format in `input_audio.format`. The response contains the transcript in `text` and the processing cost in `usage`.
Primary sources: OpenRouter, Oct 7, 2026; OpenRouter documentation and Models API, Oct 7, 2026; OpenRouter blog, Jul 22, 2026; ElevenLabs documentation, Oct 7, 2026.
The discount on all ElevenLabs models is available only on OpenRouter until Oct 19, 08:00 US Pacific Time.