1/ Audio is now first-class on OpenRouter. Two new endpoints live...
OpenRouter@OpenRouter
17 views
May 08, 2026
~1 min read
Advertisement
1
1/ Audio is now first-class on OpenRouter.
Two new endpoints live today:
📢 /api/v1/audio/speech — text-to-speech (TTS)
🎤 /api/v1/audio/transcriptions — speech-to-text (SST)
Same routing, billing, and keys you already use for text, image, and video.
Two new endpoints live today:
📢 /api/v1/audio/speech — text-to-speech (TTS)
🎤 /api/v1/audio/transcriptions — speech-to-text (SST)
Same routing, billing, and keys you already use for text, image, and video.
2
2/ Why dedicated endpoints instead of routing through chat completions?
Specialized models are dramatically faster and cheaper than general audio LLMs for the narrow jobs of "read this aloud" and "transcribe this file." Both paths are available so you can pick the right one per use case.
Specialized models are dramatically faster and cheaper than general audio LLMs for the narrow jobs of "read this aloud" and "transcribe this file." Both paths are available so you can pick the right one per use case.
3
3/ Speech launches with OpenAI GPT-4o Mini TTS, Google Gemini Flash TTS, and Mistral Voxtral Mini TTS, each with its own voice library and MP3 or PCM output. openrouter.ai/models?output_…
Transcription launches with OpenAI Whisper, GPT-4o Transcribe, Google Chirp 3, and Groq's fast Whisper inference. openrouter.ai/models?output_…
Transcription launches with OpenAI Whisper, GPT-4o Transcribe, Google Chirp 3, and Groq's fast Whisper inference. openrouter.ai/models?output_…
4
4/ The fastest way to hear it work, straight from your terminal:
5
5/ Same call with the OpenRouter TypeScript SDK:
6
6/ And in Python:
7
7/ Provider-specific knobs pass through cleanly via the provider.options field.
OpenAI speech models accept an instructions field for tone control ("speak in a warm, friendly tone").
Transcription accepts a language hint that meaningfully improves non-English accuracy. Full provider surface area, no per-vendor rewrites.
OpenAI speech models accept an instructions field for tone control ("speak in a warm, friendly tone").
Transcription accepts a language hint that meaningfully improves non-English accuracy. Full provider surface area, no per-vendor rewrites.
8
8/ Voice agents are about to be everywhere, and the supply side is diversifying fast across OpenAI, Google, Mistral, Groq, and more coming.
One API, one bill, automatic fallbacks, easy observability. Announcement: openrouter.ai/announcements/…
One API, one bill, automatic fallbacks, easy observability. Announcement: openrouter.ai/announcements/…
9
