Fish Audio models now available on Vercel AI Gateway for free
Vercel’s AI Gateway now offers free access to Fish Audio’s text-to-speech and transcription models until September 18, with usage billed afterward unless the -free suffix is added.
Vercel’s AI Gateway has integrated Fish Audio’s suite of audio models, providing developers with tools for text-to-speech and transcription. The announcement includes four models: s2.1-pro for low-latency streaming with voice cloning, transcribe-1 for timestamped text extraction, s2-pro for multilingual speech synthesis with inline style tags, and s1 for expressive text-to-speech with emotion and sound effect markers. The integration is available immediately in AI SDK 7.
To promote adoption, Vercel is waiving fees for all Fish Audio models through September 18. Standard model names will incur charges after the promotional period unless the -free suffix is appended, which prevents billing by disabling the model post-offer. Developers must explicitly opt into the free tier to avoid unexpected costs.
The models support multiple input formats for transcription, including audio buffers, base64 strings, or URLs, and return structured output with word-level timestamps. Fish Audio’s s2-pro and s2.1-pro enable fine-grained control over speech delivery via inline tags, while s1 allows dynamic adjustments for tone and sound effects. The integration aligns with Vercel’s broader AI Gateway initiative to streamline access to third-party AI services.
Users can test the models without coding by accessing the AI Gateway interface, where text or audio inputs can be processed directly in a browser. Documentation and quickstart guides are provided for implementation, including speech and transcription workflows. Fish Audio’s models are now listed alongside other providers in the AI Gateway, expanding the platform’s capabilities for real-time audio processing.