# S2.1 Pro

Fish Audio's recommended production voice model. S2.1 Pro covers 83 languages, takes direction in plain language rather than from a fixed tag list, and can render a two-speaker conversation in one pass.

- **Model ID:** `fish-audio/s2.1-pro`
- **Type:** speech
- **Providers:** fish-audio
- **Context window:** Not applicable
- **Maximum output tokens:** 0
- **Pricing:** $15/1M input tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/s2.1-pro

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { experimental_generateSpeech as generateSpeech } from 'ai';
import { gateway } from '@ai-sdk/gateway';
import { writeFile } from 'node:fs/promises';

const result = await generateSpeech({
  model: gateway.speechModel('fish-audio/s2.1-pro'),
  text: 'Hello from the Vercel AI Gateway!',
  // Browse voices at https://fish.audio/app/discovery
  // Open a voice, then use "Copy Model Id" in its "..." menu.
  voice: '933563129e564b19a115bedd57b7406a',
});

await writeFile('speech.mp3', result.audio.uint8Array);
```

## About

S2.1 Pro is Fish Audio's recommended production text-to-speech model and an improved version of S2 Pro, with better quality, latency, and throughput across 83 languages.

Prosody control is the distinguishing feature. You write direction inline in square brackets, and the model interprets it as description rather than matching against a fixed vocabulary, so `[whispers sweetly]` and `[laughing nervously]` both work without either being a predefined tag. Thousands of distinct tags are recognised, and phrasing outside that set still lands because the model learned the mapping from description to acoustics rather than from a lookup table.

Multi-speaker dialogue runs in a single request. You write both sides of a conversation in one text input, tag each turn with a speaker index, and get back one audio file with both voices rendered. Two speakers are supported per request.

In head-to-head listening comparisons against S2 Pro, S2.1 Pro was preferred in roughly three of every five judgements.

Integration uses the AI SDK's speech generation function.

## What to consider

Dialogue is capped at two speakers per request. A three-way conversation needs separate requests stitched together, which means you handle the timing between turns yourself.

Fish Audio also publishes a free development tier serving the same model quality and language coverage, without latency or availability guarantees. Use that for evaluation and prototyping, and this model where a production workload needs those guarantees.

Bracket direction is interpreted, not parsed. That flexibility is the point, but it means the same tag can land slightly differently across generations. Where a specific delivery matters, generate a few takes and pick, rather than assuming one tag produces one fixed result.

## When to use

### Best For

- **Production voice workloads** needing latency and availability guarantees
- **Multilingual narration** across 83 languages from one model
- **Expressive direction** written in plain language instead of a fixed tag vocabulary
- **Two-speaker dialogue** rendered into a single audio file in one request
- **Upgrades from S2 Pro** where listening comparisons favour this model

### Consider Alternatives When

- **Evaluation and prototyping** are the use case, where the free development tier serves the same quality
- **Three or more speakers** appear in one piece of dialogue, which exceeds the per-request cap
- **Open weights** are required, where S2 Pro is the open-source generation
- **Fine-grained emotional presets** suit the workflow better, where S1 exposes a fixed expression vocabulary

## Best for

- **Production Voice Workloads:** Latency and availability guarantees included
- **Multilingual Narration:** 83 languages from a single model
- **Expressive Direction:** Plain-language control instead of a fixed tag list
- **Two-Speaker Dialogue:** One request, one audio file, both voices
- **Upgrades From S2 Pro:** Favoured in head-to-head listening comparisons

## Consider alternatives

- **Evaluation And Prototyping:** The free development tier serves the same quality
- **Three Or More Speakers:** Exceeds the two-speaker per-request cap
- **Open Weights Requirement:** S2 Pro is the open-source generation
- **Fixed Emotional Presets:** S1 exposes an explicit expression vocabulary

## Frequently asked questions

### What is S2.1 Pro built for?

Production text-to-speech. Fish Audio recommends it as its production model, with better quality, latency, and throughput than S2 Pro.

### How many languages does S2.1 Pro support?

83.

### How do I control emotion and delivery?

Write direction inline in square brackets, such as `[whispers sweetly]`. The model interprets the description rather than matching a fixed tag list, so phrasing outside the documented set still works.

### Can S2.1 Pro render a conversation?

Yes. Write both sides in one text input, tag each turn with a speaker index, and the model returns a single audio file with both voices. Two speakers per request.

### How does S2.1 Pro compare to S2 Pro?

It is the improved version, preferred in roughly three of every five head-to-head listening judgements, with better latency and throughput and three more supported languages.

### Is there a free way to try S2.1 Pro?

Yes. Fish Audio publishes a free development tier with the same model quality and language coverage, without latency or availability guarantees. Use it for evaluation and prototyping.

### Does S2.1 Pro support Zero Data Retention?

Zero Data Retention is not currently available for this model. Zero Data Retention is offered on a per-provider basis. See https://vercel.com/docs/ai-gateway/capabilities/zdr for details.

## Links

- [Model page](https://vercel.com/ai-gateway/models/s2.1-pro)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://docs.fish.audio/developer-guide/models-pricing/models-overview)
- [Provider pricing](https://docs.fish.audio/developer-guide/models-pricing/models-overview)


If you encounter incorrect information, can’t find what you need, need a capability Vercel doesn’t currently offer, or have any other feedback to report, please do it by following the instructions at https://vercel.com/feedback.md.