Back to Blog
Model ComparisonNovember 16, 202611 min read

Songs, beats, voices, transcripts. One account.

6 music models, 15 voices, 5 transcription models on one unbottled.ai balance. The real price per clip, per minute, per hour. And which one fits which job.

A music app. A voice app. A transcription app. Three subscriptions, three logins, three piles of credits that expire. All you wanted was a jingle, a voiceover and a transcript.

unbottled.ai runs 26 audio models from 10 labs on one credit balance. Every number below comes live from our model registry: lengths, voices, tiers, real prices. Four jobs, four answers:

  • Songs with vocals: Sonauto, ElevenLabs Music, Lyria 3 Pro. Clips up to 30 seconds.
  • Beats and instrumentals: MusicGen Stereo, ModelsLab Music Gen, Lyria 3.
  • Voiceovers and read-aloud: 15 voices, $0.002 to $0.035 per minute of speech.
  • Transcription: 5 models, $0.039 to $1.06 per hour of audio.

AI music generator: all 6 models

Cheapest clip first. Prices include our 30% markup. 1 credit = $0.01.

ModelClip lengthTierCredits / clipPrice / clip
Lyria 3 Clip PreviewGoogle8 / 20 sMainstream5.2$0.05
Lyria 3Google30 sMainstream6.5$0.07
Sonauto SongSonauto8 / 20 / 30 sOpen10.4$0.10
Lyria 3 Pro PreviewGoogle30 sMainstream10.4$0.10
ModelsLab Music GenModelsLab30 sOpen27.3$0.27
ElevenLabs MusicElevenLabs8 / 20 / 30 sMainstream27.3$0.27

One price per clip, whatever length you pick. Ask for a length the model does not offer and the request snaps to the closest one it does.

Songs with vocals

A singer and a hook. Describe the genre, the mood, the voice and what the song is about. These models are built for songs that sing.

Beats and instrumentals

Background music, loops for a video, a beat to rap over. No vocals. Name genre, tempo and instruments: "lo-fi hip hop, 85 bpm, dusty piano, vinyl crackle".

Short clips, stings, jingles

Intros, notification sounds, a podcast sting. Short and cheap. Make ten, keep one.

Looking for a Suno alternative? Suno and Udio are not on unbottled.ai. What you get instead: 6 music models on one balance, from $0.05 a clip, no extra subscription. The honest catch: clips stop at 30 seconds. Full three-minute tracks still belong to the dedicated song apps.
Open the music generator

AI text to speech: 15 voices compared

Cheapest first. Billed per character of text. The last column estimates one minute of speech at about 900 characters.

VoiceSoundTierCredits / 1,000 charsAbout 1 min of speech
Bella (Kokoro)HexgradFemale, warmUncensored0.22$0.002
Sarah (Kokoro)HexgradFemale, clearUncensored0.22$0.002
Nicole (Kokoro)HexgradFemale, softUncensored0.22$0.002
Adam (Kokoro)HexgradMale, steadyUncensored0.22$0.002
Michael (Kokoro)HexgradMale, confidentUncensored0.22$0.002
Emma (Kokoro)HexgradFemale, BritishUncensored0.22$0.002
George (Kokoro)HexgradMale, BritishUncensored0.22$0.002
Ashley (Inworld)InworldFemale, livelyMainstream1.3$0.012
Craig (Inworld)InworldMale, BritishMainstream1.3$0.012
AlloyOpenAINeutral, conversationalMainstream2$0.018
NovaOpenAIFemale, warmMainstream2$0.018
OnyxOpenAIMale, deepMainstream2$0.018
Eve (xAI)xAIFemale, expressiveMainstream2.1$0.019
Zephyr (Gemini)GoogleFemale, brightMainstream3.9$0.035
Puck (Gemini)GoogleMale, upbeatMainstream3.9$0.035

Rates include our 30% markup. The minute is an estimate: fast talkers and long words shift it. You pay for the characters you send, not the seconds you get back.

Everyday read-aloud

Clean, clear, cheap enough to leave on. Nova is the default voice for reading answers aloud.

  • Nova2 credits / 1,000 chars 路 about $0.018 per minute 路 Female, warm
  • Alloy2 credits / 1,000 chars 路 about $0.018 per minute 路 Neutral, conversational
  • Onyx2 credits / 1,000 chars 路 about $0.018 per minute 路 Male, deep

Many languages, one voice

Gemini voices speak 100+ languages and follow delivery instructions: slower, whispered, excited.

  • Zephyr (Gemini)3.9 credits / 1,000 chars 路 about $0.035 per minute 路 Female, bright
  • Puck (Gemini)3.9 credits / 1,000 chars 路 about $0.035 per minute 路 Male, upbeat

Characters and storytelling

More drama, more range. Built for fiction, game characters, roleplay scenes.

  • Eve (xAI)2.1 credits / 1,000 chars 路 about $0.019 per minute 路 Female, expressive
  • Ashley (Inworld)1.3 credits / 1,000 chars 路 about $0.012 per minute 路 Female, lively

Cheapest, no content filter

Kokoro is an open-weight 82M model. 7 Kokoro voices sit in our uncensored tier: no content filter on what they read. They are also the cheapest voices on the platform.

  • Bella (Kokoro)0.22 credits / 1,000 chars 路 about $0.002 per minute 路 Female, warm
  • Michael (Kokoro)0.22 credits / 1,000 chars 路 about $0.002 per minute 路 Male, confident
  • Emma (Kokoro)0.22 credits / 1,000 chars 路 about $0.002 per minute 路 Female, British
  • George (Kokoro)0.22 credits / 1,000 chars 路 about $0.002 per minute 路 Male, British

Every voice here is tuned for English. Gemini voices handle 100+ languages. The rest can read other languages, often with an accent. Test one sentence before you render a whole script.

Open text to speech

AI transcription: 5 speech to text models

Cheapest per minute first. Billed per second of audio. Prices include our 30% markup.

ModelTierCredits / minutePrice / hour
Whisper Large V3OpenAIOpen0.064$0.039
Voxtral Mini TranscribeMistral AIOpen0.5$0.30
GPT-4o Mini TranscribeOpenAIMainstream0.66$0.40
WhisperOpenAIMainstream0.78$0.47
Nemotron ASRNVIDIAOpen1.8$1.06

When a provider reports its actual cost for a job, we bill that cost plus the markup. A final charge can land slightly off the table.

Cheapest: hours of audio for cents

Open-weight Whisper, 99 languages. One subscription covers about 207 hours of audio at this rate.

Accuracy first

Fewer wrong words, better language detection. For interviews, quotes, anything you publish. Whisper is the default for voice input.

Many languages

Whisper covers 99 languages, Nemotron 40 locales. Voxtral is Mistral's transcription model, built for recordings up to 10 minutes.

Fast and noise-robust

Low latency, holds up in loud rooms. Made for quick voice notes and dictation.

Open speech to text

How it works in the chat

No separate app. Music, voice and transcription live inside every chat on unbottled.ai.

1

Ask for a song

Type "make a 30 second synthwave intro for my podcast". The AI calls the music generator with your skill's music model, Lyria 3 unless you changed it. Name another model in the prompt to switch.

2

Hear any answer

Hit the speaker icon on a reply and it is read aloud in your skill's voice, Nova by default. Set a different voice per skill.

3

Talk instead of typing

Tap the mic and speak. Whisper transcribes your words into the input, or the recording goes straight to models that understand audio.

4

Transcribe whole files

Meetings, interviews, voice memos: drop up to 20 files on the speech to text tool page and pick the model.

5

Call mode

Hands-free. You speak, the AI answers out loud, you answer back. Voice in, voice out, same balance.

Want full control? Each job has its own tool page: pick the model or voice, set the clip length, paste a long script.

What AI music and voice cost here

Three formulas, one balance. Provider price plus our 30% markup. 1 credit = $0.01.

  • Music: a flat price per clip. From 5.2 credits ($0.05) on Lyria 3 Clip Preview to 27.3 credits ($0.27) on ElevenLabs Music.
  • Voice: per character. From 0.22 credits per 1,000 characters on Bella (Kokoro) to 3.9 on Puck (Gemini).
  • Transcription: per second of audio. From 0.064 credits a minute on Whisper Large V3 to 1.8 on Nemotron ASR.
  • Credits come off only once you have the result. Failed job, timeout, refused prompt: you pay nothing.
  • Free: 20 credits a month without an account, 40 with a free account, 90 with the newsletter. 40 credits buy 7 clips on Lyria 3 Clip Preview, or about 201 minutes of Bella (Kokoro).
  • Monthly plans: four sizes, starting at 800 credits for $8.00.
  • Credit packs: a top-up for subscribers, 500 credits for $5.00. They never expire. Card, PayPal or crypto.
See plans and credit packs

The limits, straight

Know the edges before you spend a credit.

  • Music clips run 30 seconds at most. No three-minute tracks, no extending a clip.
  • A clip costs the same at every length. Short clips are the cheap way to test a prompt, not a discount.
  • No Suno, no Udio. Their APIs are not on unbottled.ai.
  • Text to speech takes up to 5,000 characters per request. Longer scripts: split them.
  • Transcription takes up to 20 files per request, 25 MB each.
  • No voice cloning. You pick from the voices in the table. You cannot upload your own.
  • Most music and voice models run behind their provider's moderation. A prompt can be refused upstream. That refusal costs nothing.
Before you publish: check the provider's terms for commercial use of generated music and voices. Rules differ per lab, and they change.

FAQ

What is the best AI music generator?

Depends on the track. Songs with vocals: Sonauto or ElevenLabs Music. Polished short tracks: Lyria 3 Pro. Beats and loops: MusicGen Stereo. Cheapest: Lyria 3 Clip Preview at $0.05 a clip. All 6 run on one unbottled.ai balance, so you switch per clip.

Is there a good Suno alternative?

For short songs and beats, yes: Sonauto, ElevenLabs Music and Google's Lyria 3 run here on one balance, from $0.05 a clip, no extra subscription. For full three-minute songs, not yet: clips here stop at 30 seconds.

Can AI make songs with vocals?

Yes. Ask Sonauto, ElevenLabs Music or Lyria 3 Pro for a song and describe the topic, the mood and the singer. Instrumental models like MusicGen Stereo make music without a voice.

Which AI voice generator sounds most natural?

For everyday read-aloud, OpenAI voices like Nova sound clean and cost little. Most expressive: xAI Eve and Inworld. For many languages: Gemini. For the lowest price and no filter: Kokoro.

Is there an AI voice generator with free credits?

Yes. 20 free credits a month without an account, 40 with a free account, 90 with the newsletter. On Bella (Kokoro) that is about 201 minutes of speech. The same credits pay for music and transcription.

How much does AI transcription cost?

From $0.039 per hour of audio on Whisper Large V3 to $1.06 on Nemotron ASR. Billed per second. Failed jobs cost nothing.

Can I clone my own voice?

No. unbottled.ai offers 15 ready-made voices. No cloning, no uploading your own voice.

Is there an uncensored text to speech?

Yes. 7 Kokoro voices sit in our uncensored tier: an open-weight model with no content filter on what it reads. Most music models and the other voices run behind their provider's filters.

Making music, podcasts or voiceover content? The unbottled.ai partner program pays you a recurring share of every payment you refer. Join the partner program

Pick a model. Hit play.

Songs, beats, voices and transcripts from one balance. Start with 20 free credits, no account needed.

Back to Blog