What is Stable Audio 2.5?
Stable Audio 2.5 turns a text prompt, or a clip of your own audio, into a track of up to 190 seconds in 44.1 kHz stereo. It handles text-to-audio, audio-to-audio and inpainting, where you mark a stretch of a track and the model rewrites only that part. Stability trained it on licensed AudioSparx music and Freesound recordings, and says a full track takes under two seconds on a GPU.
It is no longer Stability's lead audio model. Stable Audio 3.0 now runs the Stable Audio app and plugin, while 2.5 stays on Stability's API as the cheaper call, with no retirement announced. Unlike Suno or Lyria 3.5, it takes no lyrics.
Is Stable Audio 2.5 still available?
Yes, on Stability AI's API, but Stability now leads with Stable Audio 3.0 everywhere else.
- Stability AI API
- Still listed at 20 credits per generation, with no deprecation notice in Stability's API release notes.
- Stable Audio app and plugin
- Both now run Stable Audio 3.0 models, not 2.5.
- Stability's audio page
- Lists only the Stable Audio 3.0 family: Large, Medium, Small and Small SFX.
- Successor
- Stable Audio 3.0 is Stability's newest audio model; its Large size sits on the same API at 26 credits per generation.
Who is Stable Audio 2.5 best for?
Stable Audio 2.5 is best for brands, agencies and developers who want licensed-data music and sound effects in bulk through an API, up to three minutes a track.
Brand and ad audio
Stability built it for companies that need their own sound across ads, games and stores, and its team can fine-tune it on a brand's sound library.
Music drafts at volume
At $0.20 a track against $0.26 for Stable Audio 3.0, with multi-part tracks back in seconds, batches of options stay cheap.
Reworking existing audio
Audio-to-audio and inpainting let you feed in your own recording, mark the part to change and let the model fill it in.
Commercial projects
Stability trained it on licensed data and allows commercial use of API outputs, which matters for client work.
What's included with Stable Audio 2.5?
Text-to-audio, audio-to-audio and inpainting through one API, with tracks up to 190 seconds in MP3 or WAV.
- Text-to-audio
- Prompts of up to 10,000 characters, with the length set anywhere from 1 to 190 seconds.
- Audio-to-audio
- Upload an MP3 or WAV of 6 to 190 seconds and a strength setting decides how much of it survives.
- Inpainting
- Set a start and end time and the model regenerates only that section, using the rest of the track as context.
- Multi-part structure
- Tracks come with an intro, development and outro rather than one looping idea.
- Custom fine-tuning
- For enterprise customers, Stability's audio team can tune the model on a company's own sound library.
- Several ways to run it
- Stability's API, partner platforms such as fal, Replicate and ComfyUI, or on your own servers under an Enterprise license.
How much does Stable Audio 2.5 cost?
Stable Audio 2.5 costs 20 credits, or $0.20, per track on Stability AI's API, and new API accounts get 25 free credits to start.
- Tracks up to three minutes from a text or audio prompt
- Credits cost $0.01 each, bought on your account page
- 25 free credits when you open an account
Failed generations are not charged. The newer Stable Audio 3.0 costs 26 credits ($0.26) per generation on the same API. The Stable Audio web app and DAW plugin at stableaudio.com now run Stable Audio 3.0 and are sold as monthly plans from $12 (Solo, 660 credits) to $199 (Studio, 14,000 credits). Companies that want to run Stability's audio models on their own servers need an Enterprise license, priced on request. Full pricing on platform.stability.ai. We checked these prices on .
How do readers rate Stable Audio 2.5?
No one has rated Stable Audio 2.5 yet. Be the first.
Sign in to rate. One rating per account.
Stable Audio 2.5 pros and cons
It is cheap, fast and follows prompts closely by Stability's own test, but tracks stop at about three minutes, it takes no lyrics and Stable Audio 3.0 is now the lead model.
Pros
- Costs $0.20 a track on Stability's API, less than the $0.26 Stable Audio 3.0 call.
- Scored highest for matching the prompt among the Stable Audio models in Stability's own listening test on the Song Describer Dataset: 4.30 out of 5, against 3.80 for 3.0 Large.
- Fast: Stability measures under two seconds per full track on a GPU.
- Trained on licensed AudioSparx music and Freesound recordings, and Stability allows commercial use of API outputs.
Cons
- Tracks stop at 190 seconds, while Stable Audio 3.0 Medium and Large reach six minutes twenty seconds.
- The API has no lyrics field; it makes music and sound effects, so songs with sung words need another model.
- Closed weights: there is no download, and self-hosting needs a paid Enterprise license, unlike three of the four Stable Audio 3.0 sizes.
- Stability's app, plugin and audio model pages now feature only Stable Audio 3.0, and 2.5 scored lower on musicality than 3.0 Medium and Large in Stability's own test.
What are the best Stable Audio 2.5 alternatives?
Three tools on Recaply do Stable Audio 2.5's job. ACE-Step 1.5 is open source, so you can run it on your own server.
Stable Audio 3.0FreeChoose it when you need tracks longer than three minutes, up to 6:20, or weights you can run yourself; its API call costs $0.26 instead of $0.20.
Lyria 3.5FreeChoose it when you need full songs with vocals and lyrics; Google's Gemini API charges $0.08 per song, though it has no free tier.
ACE-Step 1.5Open SourceChoose it when you want free, local generation with vocals; it is open source under the MIT licence and runs in under 4GB of VRAM.
Questions about Stable Audio 2.5
What is Stable Audio 2.5?
Stable Audio 2.5 is Stability AI's licensed-data model for music and sound effects up to three minutes long, sold per track on Stability's API and built for brands and agencies.
Who is Stable Audio 2.5 for?
Stable Audio 2.5 is best for brands, agencies and developers who want licensed-data music and sound effects in bulk through an API, up to three minutes a track.
Is Stable Audio 2.5 free?
Not beyond a first try. New Stability AI API accounts get 25 free credits, enough for one Stable Audio 2.5 track at 20 credits; after that each track costs $0.20. The stableaudio.com pricing page lists only paid plans, and that app now runs Stable Audio 3.0.
How much does Stable Audio 2.5 cost?
Stable Audio 2.5's paid plans are Stability AI API at $0.20 per generation (20 credits).
Is there an open source alternative to Stable Audio 2.5?
Yes. ACE-Step 1.5 is an open source tool that does the same job, so you can run your own.
What are the best Stable Audio 2.5 alternatives?
The closest alternatives on Recaply are Stable Audio 3.0, Lyria 3.5 and ACE-Step 1.5.
Where can I try Stable Audio 2.5?
On its official website, platform.stability.ai.
How does Stable Audio 2.5 compare with Stable Audio 3.0?
Stable Audio 2.5 makes tracks up to three minutes and costs $0.20 per generation on Stability AI's API. Stable Audio 3.0 costs $0.26 there, its Medium and Large sizes reach six minutes twenty seconds, and three of its four sizes are free to download as open weights.
Can I use Stable Audio 2.5 music commercially?
Yes. Stability AI says you can use outputs from its API for commercial purposes as long as they do not break anyone's copyright, and it trained Stable Audio 2.5 on licensed data. Audio you upload for audio-to-audio or inpainting must be free of copyrighted material.
Can I run Stable Audio 2.5 on my own computer?
No. Stability has not released Stable Audio 2.5's weights, so it runs on Stability AI's API and partner platforms. Companies can deploy it on their own servers only under a paid Enterprise license. For local generation, Stable Audio 3.0 Small and Medium are free to download.
Our verdict
Stable Audio 2.5 still works as a cheap API call for short licensed tracks, but new projects should start with Stable Audio 3.0.
It costs $0.20 a track against $0.26 and matched prompts best in Stability's own test, yet 3.0 Medium and Large make tracks twice as long, score higher on musicality and come with open weights.
- Skip it for
- Anyone who needs vocals, tracks longer than about three minutes, or a model they can run offline.
- Check first
- Run the same prompts through Stable Audio 3.0 on the same API before you commit.
Lyria 3
Magenta RealTime 2
Suno v6
Songtell
Musick.ai
AI Music Maker