Supported Models

Updated 2026-09-18 · Suno-API team

Currently available generation models

Customer-facing requests use only the platform model names in the table below.

Default recommendation: suno-v6.

A note on stems

Stem separation is a standalone fixed capability; it does not require you to pick a song-generation model. When you call the stem endpoint you only supply the source song and the stem mode, for example two or twelve. The number of results actually returned is whatever the response reports.

The model parameter

In generation endpoints model is usually optional:

• Omitted: the current default suno-v6 is used.
• Supplied: it must be suno-v6, suno-v6-wild or suno-v6-mini.
• Only the platform model names in the table above are accepted.

Minimal generation example

curl -X POST https://www.suno-api.io/api/music/create \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  --max-time 300 \
  -d '{
    "description": "an upbeat pop song about the first day of spring",
    "model": "suno-v6"
  }'

Waiting time and timeouts. Generation endpoints are synchronous: the order is only accepted once the response arrives. Measured on production, most requests return in 1-3 seconds, about 10% take longer than 30 seconds, and the longest observed wait is about 230 seconds. Set your client timeout to 300 seconds or more (the --max-time 300 above); a timeout does not mean the song was not generated, so do not retry immediately - check GET /api/music/songs first.

V6 variant example

{
  "model": "suno-v6-wild",
  "description": "an electronic track with an experimental structure"
}

Retired models

The following models can no longer be used for new requests:

• suno-v2
• suno-v3
• suno-v3-5
• suno-v4
• suno-v4-5
• suno-v4-5-all
• suno-v4-5-plus
• suno-v5
• suno-v5-5

Existing songs and historical tasks may still report an old model_name or model_version in query results. That is historical data; it does not mean the old model can accept new requests.

For compatibility with older clients that are still running, a request that names a retired model is processed with the current default suno-v6 instead of being rejected. The task record will report suno-v6; you will not receive a "model unavailable" error. You should still update your request examples to the current model names as soon as possible.

Listing models

GET /v1/models should include at least these standard song models:

{
  "object": "list",
  "data": [
    { "id": "suno-v6", "object": "model", "owned_by": "suno" },
    { "id": "suno-v6-wild", "object": "model", "owned_by": "suno" },
    { "id": "suno-v6-mini", "object": "model", "owned_by": "suno" }
  ]
}

The response may also contain lyric models, compatibility routes or other internal models. Do not confuse those with the V6 song-generation models.

Reading model fields in responses

model_version, model and model_name in query results can come from either the stored platform value or the upstream response:

• The stored current generation model normally reads suno-v6, suno-v6-wild or suno-v6-mini.
• The model field returned by the API always uses the platform model name.
• Historical tasks may show a retired model name. Treat that as historical display data and do not resubmit requests with it.

Choosing between the V6 variants

Most integrations only ever need the default. The three variants differ in how tightly they follow the prompt versus how much variation they introduce:

Because a run returns two tracks either way, a practical workflow is to draft with suno-v6-mini, then re-run the prompt you liked on suno-v6 for the take you actually publish.

Cost and refunds

Generation is billed per request, not per model: each run costs 0.6 CNY and returns two tracks. A failed generation is refunded automatically, so an unsuccessful attempt never consumes balance. Downloads are unlimited and are never billed separately.