Do I have to rewrite my code?

No. Fish Audio's OpenAI-compatible API accepts the same /v1/audio/speech request from the official Python and Node SDKs. Change base_url and api_key, set a Fish model, and pass a Fish voice ID.

Is fish-audio/s2.1-pro-free really free?

Yes, through November 30, 2026, under fair-use limits. It is the same model as S2.1-Pro, without guarantees on time to first audio (TTFA) or a DPA, and requests may be used to improve model quality. Use it to test and prototype, then switch to fish-audio/s2.1-pro for production.

Does speech-to-text move the same way?

Yes. /v1/audio/transcriptions accepts whisper-1, gpt-4o-transcribe and gpt-4o-mini-transcribe as aliases for fish-audio/transcribe-1. Chat-completions audio and a Realtime WebSocket are supported too; /v1/audio/translations is not.

What does not carry over?

instructions, the sse stream format and the aac and flac output formats return a 400. OpenAI preset voice names are not available, and the compatible API does not list Fish voices; browse them in the Voice Library.

OpenAI removes tts-1, tts-1-hd and gpt-4o-mini-tts from its API on January 6, 2027.Source

OpenAI TTS alternative: keep the SDK, change the base URL

Fish Audio serves the same /v1/audio/speech call. Point your OpenAI client at Fish, pick the free model, and your speech code keeps running after OpenAI's TTS models are gone.

Get your free API key Get your free API key Read the migration guide Read the migration guide

No credit card required: fish-audio/s2.1-pro-free costs $0 through November 30, 2026.

Two lines move the client

base_url and api_key point the official OpenAI SDK at Fish. In the call, set a Fish model and voice. The new code runs as-is on a fresh account with a $0 balance.

speech.py

import os from openai import OpenAI client = OpenAI(+Added line:     base_url="https://api.fish.audio/compat/v1",-Removed line:     api_key=os.environ["OPENAI_API_KEY"],+Added line:     api_key=os.environ["FISH_API_KEY"],) with client.audio.speech.with_streaming_response.create(-Removed line:     model="tts-1",-Removed line:     voice="alloy",+Added line:     model="fish-audio/s2.1-pro-free",+Added line:     voice="",    input="Hello, world!",    response_format="mp3",) as response:    response.stream_to_file("speech.mp3")

import fs from "node:fs";import OpenAI from "openai"; const client = new OpenAI({+Added line:   baseURL: "https://api.fish.audio/compat/v1",-Removed line:   apiKey: process.env.OPENAI_API_KEY,+Added line:   apiKey: process.env.FISH_API_KEY,}); const response = await client.audio.speech.create({-Removed line:   model: "tts-1",-Removed line:   voice: "alloy",+Added line:   model: "fish-audio/s2.1-pro-free",+Added line:   voice: "",  input: "Hello, world!",  response_format: "mp3",});fs.writeFileSync("speech.mp3", Buffer.from(await response.arrayBuffer()));

-Removed line: curl https://api.openai.com/v1/audio/speech \-Removed line:   -H "Authorization: Bearer $OPENAI_API_KEY" \+Added line: curl https://api.fish.audio/compat/v1/audio/speech \+Added line:   -H "Authorization: Bearer $FISH_API_KEY" \  -H "Content-Type: application/json" \  -d '{-Removed line:     "model": "tts-1",-Removed line:     "voice": "alloy",+Added line:     "model": "fish-audio/s2.1-pro-free",+Added line:     "voice": "",    "input": "Hello, world!",    "response_format": "mp3"  }' \  --output speech.mp3

Needs the official OpenAI SDK (pip install openai or npm install openai) and your Fish API key in FISH_API_KEY.

What else changes

voice takes a Fish voice ID. OpenAI preset names such as nova return a 400; an empty string uses the model's default voice.
response_format defaults to pcm on Fish, not mp3. Set it explicitly.
instructions is not supported and returns a 400. Write delivery cues into the text instead, such as [whispering].
tts-1, tts-1-hd and gpt-4o-mini-tts still work as model names. They map to the paid fish-audio/s2.1-pro.

Pick a voice

OpenAI's preset voices don't exist on Fish. Pass the ID of any voice in the Fish Voice Library, or of a voice you clone. An empty string uses the model's default voice.

Hannah · Voice Library

voice="9a9cf47702da476aa4629e2506d4a857"

Browse the Voice Library

Pricing

S2.1-Pro bills by UTF-8 bytes of input text. In English one character is one byte, so it costs the same as tts-1 and half as much as tts-1-hd. Most Chinese, Japanese and Korean characters take 3 bytes.

Model	Price	Notes
Fish Audiofish-audio/s2.1-pro-freeFree through November 30, 2026, under fair-use limits. Same model as S2.1-Pro, without latency (TTFA) or DPA guarantees; requests may be used to improve the model.	$0.00 / 1M UTF-8 bytes	Free through November 30, 2026, under fair-use limits. Same model as S2.1-Pro, without latency (TTFA) or DPA guarantees; requests may be used to improve the model.
Fish Audiofish-audio/s2.1-proFor production. Pay as you go, no subscription.	$15.00 / 1M UTF-8 bytes	For production. Pay as you go, no subscription.
OpenAItts-1Removed from the OpenAI API on January 6, 2027.	$15.00 / 1M characters	Removed from the OpenAI API on January 6, 2027.
OpenAItts-1-hdRemoved from the OpenAI API on January 6, 2027.	$30.00 / 1M characters	Removed from the OpenAI API on January 6, 2027.

Prices from the official pricing pages, checked October 29, 2026:Fish Audio OpenAI

Frequently asked questions

OpenAI's deprecations page lists tts-1, tts-1-hd and gpt-4o-mini-tts for removal from the API on January 6, 2027. Its recommended replacement, gpt-realtime-2.1-mini, runs on the Realtime API rather than /v1/audio/speech.

Switch before January 6, 2027

Create a free key, change two lines, and keep shipping speech.