Last verified: August 20, 2026
Hosted API access update — August 20, 2026: MiniMax says paid Music Generation and Lyrics Generation APIs are no longer available to new users; existing paying users can continue using their current API services. The free IDs
music-3.0-free,music-2.6-free, andmusic-cover-freeare discontinued. New users are directed to MiniMax Audio or the downloadable MiniMax Music 3 checkpoint. The checkpoint is a separate open-weight route under its Community License and is not the hosted Lyrics or Music Cover API. Some operation schemas still display the former free IDs, so schema visibility must not be treated as current entitlement.
Previous-generation model notice: MiniMax now recommends
music-3.0for text-to-music generation.music-2.6remains documented for eligible existing paying users, whilemusic-2.6-freeis discontinued as of August 20, 2026. Use this page for Music 2.6-specific integrations, then review the MiniMax Music 3.0 guide before starting a new project.
MiniMax Music 2.6 is a text-to-music model for generating vocal songs or instrumental music from a written brief. It supports structured lyrics, automatic lyric generation, streaming or non-streaming output, and configurable audio files through the official MiniMax Music Generation API.
This independent guide explains the verified model IDs, request fields, pricing, limits, prompt workflow, and migration path. It separates the music-2.6 text-to-music model from the distinct music-cover route, an important distinction when adapting older examples.
Independent reference: MiniMax-AI.chat is not affiliated with MiniMax and does not provide API keys, billing, storage, or music-generation services. The official console, API schema, account terms, and generated response control your use.
MiniMax Music 2.6 Guide Contents
- Verified status, model IDs, price, and rate limits
- What Music 2.6 is and what MiniMax announced
- Music 2.6 compared with Music 3.0
- Secure Python API example
- Request fields and character limits
- Audio formats, streaming, and response handling
- Prompt and lyrics guide
- Why Music Cover is a separate model
- Pricing and rate limits
- Use cases, limitations, and migration decisions
- Rights and responsible-use checklist
- Frequently asked questions
- Official sources and verification notes
MiniMax Music 2.6 Verified Snapshot
| Item | Verified value | What it means |
|---|---|---|
| Status | Previous-generation text-to-music model | Still listed by the API, but MiniMax recommends Music 3.0 |
| Paid model ID | music-2.6 | Eligible existing paying users only; new API access is closed |
| Free model ID | music-2.6-free | Discontinued on August 20, 2026 |
| API endpoint | POST https://api.minimax.io/v1/music_generation | Official text-to-music generation route |
| Authentication | Authorization: Bearer API_KEY | Keep the key on a trusted server |
| Paid price | $0.15 per up-to-5 minutes of music | Official pay-as-you-go billing unit; verify before budgeting |
| Paid rate limit | 120 RPM | Official model description; account limits can still vary |
| Free route | Discontinued | The schema still shows a historical 3-RPM value, but MiniMax’s August 20 notice controls current availability |
| Launch announcement | April 10, 2026 | MiniMax’s dedicated Music 2.6 release article |
The exact API strings are lowercase. An eligible existing paying account uses music-2.6; music-2.6-free remains visible in the schema but is discontinued. A third-party provider may invent a namespace-style identifier; that provider-specific name is not the official MiniMax model ID.
What Is MiniMax Music 2.6?
Music 2.6 turns a style description and, when required, lyrics into generated audio. The prompt can describe genre, mood, intended scene, instruments, vocal character, tempo, key, song structure, and emotional development. For a vocal song, lyrics provide the words and section map. For instrumental generation, the prompt becomes required and the lyrics field can be omitted.
At launch, MiniMax highlighted stronger instruction control, redesigned mid-to-low-frequency acoustics, and better handling of tension, pauses, and dramatic progression. Its release article also reported first-packet latency below 20 seconds. These are MiniMax’s dated launch claims, not a latency guarantee for every account, region, network, prompt, or output mode.
Instruction and Arrangement Direction
MiniMax said Music 2.6 improved its response to BPM, key, structure, and emotional-arc instructions. These values are normally written into the natural-language prompt; they are not separate bpm or key properties in the official Music Generation request schema.
Vocal and Instrumental Generation
The current API supports both vocal and instrumental Music 2.6 requests. A vocal request normally supplies structured lyrics. An instrumental request sets is_instrumental to true, requires a descriptive prompt, and does not require lyrics. This makes the same model useful for song drafts, background cues, game concepts, podcast beds, and other non-vocal work.
Automatic Lyrics
When lyrics_optimizer is true and the lyrics value is empty, MiniMax documents automatic lyric generation from the prompt. Despite the property name, the public schema describes this empty-lyrics workflow; it does not document the field as an editor for an existing lyric sheet. MiniMax also lists a separate Lyrics Generation/Editing service at $0.01 per song.
MiniMax Music 2.6 vs Music 3.0
MiniMax released Music 3.0 on July 16, 2026 and labels it the recommended text-to-music model. The two generations share the same hosted endpoint and core request fields for existing entitled accounts, but both free IDs are discontinued and new paid API access is closed. The main decision is therefore model generation and output preference, not a completely different integration architecture.
| Comparison | Music 2.6 | Music 3.0 |
|---|---|---|
| Official status | Previous generation | Recommended generation |
| Paid ID | music-2.6 | music-3.0 |
| Free ID | music-2.6-free | music-3.0-free |
| Paid price | $0.15 per up-to-5 minutes | $0.15 per up-to-5 minutes |
| Hosted access | Existing paying accounts only; free ID discontinued | Existing paying accounts only; free ID discontinued |
| Official highlights | Instruction control, low-end response, tension, and Cover release family | Intent understanding, sound quality, instrument technique, and more natural vocal synthesis |
| Best starting point | Maintaining or reproducing a 2.6-specific workflow | Eligible existing hosted-account migration, or evaluation of MiniMax Audio and the separate open-weight Music 3 route |
Do not assume a newer model will improve every prompt in exactly the same way. If an application already depends on 2.6, run a controlled migration: send the same licensed prompts and lyrics to both models, save outputs and request settings, assess them with a documented rubric, and change the model ID only after the results meet the project’s requirements. See the dedicated Music 3.0 features and API guide for the newer release.
MiniMax Music 2.6 API Quick Start
Create an API Key through the official MiniMax platform and store it in a server-side environment variable. Never place a production key in front-end JavaScript, a mobile bundle, a screenshot, a public repository, or an article example.
export MINIMAX_API_KEY="replace-with-your-server-side-key"
pip install requestsThe following Python example sends original sample lyrics to the official endpoint and requests a temporary URL instead of printing a large hexadecimal audio value:
import os
import requests
api_key = os.environ["MINIMAX_API_KEY"]
url = "https://api.minimax.io/v1/music_generation"
headers = {
"Authorization": f"Bearer {api_key}",
"Content-Type": "application/json",
}
payload = {
"model": "music-2.6",
"prompt": (
"Warm indie folk, 92 BPM, reflective mood, intimate lead vocal, "
"fingerpicked acoustic guitar, soft piano, restrained drums, "
"and a gentle lift in the final chorus"
),
"lyrics": (
"[Verse]\n"
"Morning gathers on the glass\n"
"Quiet footsteps let it pass\n\n"
"[Chorus]\n"
"Carry the light into the day\n"
"Leave a little room along the way"
),
"stream": False,
"output_format": "url",
"audio_setting": {
"sample_rate": 44100,
"bitrate": 256000,
"format": "mp3",
},
}
response = requests.post(
url,
headers=headers,
json=payload,
timeout=180,
)
response.raise_for_status()
result = response.json()
base_resp = result.get("base_resp", {})
if base_resp.get("status_code") != 0:
raise RuntimeError(base_resp.get("status_msg", "MiniMax API error"))
data = result.get("data", {})
if data.get("status") != 2:
raise RuntimeError(f"Generation did not complete: {data.get('status')}")
audio_url = data.get("audio")
if not audio_url:
raise RuntimeError("The response did not include an audio URL")
print("Audio URL (download before it expires):", audio_url)MiniMax states that URL outputs expire after 24 hours, so a production service should download the file promptly to authorized storage rather than treating the returned URL as a permanent media address. Add application-specific retries, content review, logging that excludes secrets, request cancellation, and storage retention rules before deployment.
Do not start a new integration with music-2.6-free or music-3.0-free: MiniMax says both free routes are discontinued. The example above remains relevant only to an eligible existing paying account using music-2.6. For broader authentication and error-handling guidance, use the MiniMax API guide.
Music 2.6 Request Fields and Limits
| Field | Music 2.6 rule | Practical note |
|---|---|---|
model | Required; use music-2.6 only if the account already has access | music-2.6-free is discontinued |
prompt | Instrumental: required, 1–2,000 characters. Vocal: optional, 0–2,000 | Describe style, mood, scenario, instruments, and direction |
lyrics | Instrumental: not required. Vocal: required, 1–3,500 characters | Separate lines with \n and use supported section tags |
stream | Optional boolean; default false | Streaming output supports only hexadecimal audio |
output_format | url or hex; default hex | URL outputs expire after 24 hours |
audio_setting | Optional audio configuration object | Choose a documented sample rate, bitrate, and file format |
lyrics_optimizer | Optional boolean; default false | With empty lyrics, generates lyrics from the prompt |
is_instrumental | Optional boolean; default false | Set true for music without vocals |
Validation depends on the mode. A non-instrumental request cannot simply omit meaningful lyric input unless it uses the documented automatic-lyrics workflow. An instrumental request should not rely on empty lyrics alone; explicitly set is_instrumental to true and provide a specific prompt.
Audio Settings, Output, and Errors
| Setting | Official options | Selection note |
|---|---|---|
| Sample rate | 16000, 24000, 32000, 44100 | Use 44.1 kHz when a standard music-delivery rate fits the workflow |
| Bitrate | 32000, 64000, 128000, 256000 | Higher bitrate usually increases file size |
| Audio format | mp3, wav, pcm | Choose for delivery, editing, or raw processing requirements |
| Response representation | url or hex | Streaming requires hex; non-streaming can request either |
The response data uses status 1 for in progress and 2 for completed. A successful HTTP response does not by itself prove that generation succeeded: inspect base_resp.status_code. Published codes include success 0, rate limit 1002, authentication failure 1004, insufficient balance 1008, sensitive-content flag 1026, invalid parameters 2013, and invalid API Key 2049.
- Retry temporary rate-limit or network failures with bounded exponential backoff and jitter.
- Do not retry invalid credentials, insufficient balance, invalid fields, or flagged content without correcting the cause.
- Record the model ID, prompt version, response status, trace identifier, and cost context without logging the API key.
- Validate downloaded content type and size before making the file available to an application user.
How to Prompt MiniMax Music 2.6
Treat the prompt as a compact production brief. A practical order is: intended use, genre, mood, tempo or BPM, key if important, vocal direction, lead instruments, rhythm section, arrangement movement, mix character, and ending. Include only details that matter; conflicting genres, tempos, or emotional directions make the target less coherent.
Vocal Song Prompt
D major, 112 BPM, uplifting indie pop for a product story, clear warm lead vocal, bright guitar, rounded bass, tight drums, restrained verses, a wider pre-chorus, memorable final chorus, clean modern mix, decisive ending.Describe vocal qualities rather than requesting an imitation of a real singer. Terms such as “warm alto,” “restrained delivery,” “conversational phrasing,” or “layered harmony in the final chorus” express a production goal without asking the model to copy a recognizable person.
Instrumental Prompt
E minor, 96 BPM, cinematic instrumental for a mystery-game menu, low piano pulse, bowed strings, distant metallic percussion, gradual tension, no vocals, loop-conscious ending with a short reverb tail.For an instrumental request, also set "is_instrumental": true. Writing “no vocals” in the prompt can reinforce the creative direction, but it should not replace the documented boolean.
Lyrics and Structure Tags
The official schema supports [Intro], [Verse], [Pre Chorus], [Chorus], [Interlude], [Bridge], [Outro], [Post Chorus], [Transition], [Break], [Hook], [Build Up], [Inst], and [Solo]. Use tags to communicate function, not to overload every line with production notes.
- Keep lines singable and revise crowded wording before generating multiple variations.
- Place the main lyrical idea in the chorus and use verses for development.
- Use a bridge or interlude only when the song benefits from contrast.
- Make arrangement direction in the prompt consistent with the section map in the lyrics.
- Use original or properly licensed lyrics and keep a record of their source.
Music Cover Is a Separate Model Route
Cover was introduced as part of the Music 2.6 release story, but the hosted API does not perform reference-audio Cover work by setting model to music-2.6. An eligible existing paying account uses music-cover; music-cover-free is discontinued. This separation matters for fields, validation, rights review, and rate limits.
| Cover item | Official rule |
|---|---|
| One-step input | Provide exactly one of audio_url or audio_base64 plus a target-style prompt |
| Reference duration | 6 seconds to 6 minutes |
| Reference size | Maximum 50 MB |
| Reference format | Common audio formats such as MP3, WAV, or FLAC |
| Cover prompt | Required, 10–300 characters |
| Cover lyrics | Optional for one-step Cover; if omitted, the service extracts them through ASR; 10–1,000 characters when supplied |
| Two-step workflow | Preprocess the audio, edit structured lyrics, then submit a cover_feature_id |
| Feature ID lifetime | 24 hours |
A reference recording, composition, lyrics, or voice may carry rights that are independent of the model. Technical acceptance by the API does not establish permission to upload, transform, publish, or monetize it.
Music 2.6 Pricing and Rate Limits
| Service | Published price | Published rate |
|---|---|---|
music-2.6 | $0.15 legacy list price for eligible existing paying accounts | 120 RPM remains the published technical ceiling for entitled accounts |
music-2.6-free | Discontinued | Historical schema value: 3 RPM; not current availability |
| Lyrics Generation | $0.01 per song | Check the account and endpoint documentation |
MiniMax’s general rate-limit table groups Music 3.0, Music 2.6, Music Cover, and legacy Music 2.0 at 120 requests per minute and 20 concurrent connections for entitled accounts. The endpoint schema still identifies the discontinued free variants at a historical 3 RPM. These rows describe technical ceilings, not access for a new account.
The phrase “per up-to-5 minutes of music” is the published billing unit. It should not be rewritten as a guarantee that every prompt produces exactly five minutes. Estimate batch cost as successful billable generations multiplied by the applicable unit rate, then allow for revisions, rejected takes, lyric calls, storage, transfer, and any third-party platform fees. Check the live official table and the independent MiniMax pricing reference before committing a budget.
Best Uses, Limitations, and Migration Decisions
Where Music 2.6 Can Fit
- Existing integrations: maintaining a known 2.6 model ID while a team evaluates Music 3.0.
- Song and arrangement drafts: turning original lyrics and a production brief into material for review.
- Instrumental concepts: exploring cues for videos, games, presentations, podcasts, or prototypes.
- Structured creative experiments: testing how BPM, key, sections, instruments, and emotional direction affect results.
- Backend creative tools: generating audio through a controlled server-side workflow with authorization, logging, and cost limits.
Limits to Plan Around
Generated results can vary between requests. A track may need prompt revision, lyric revision, trimming, loudness work, arrangement edits, mixing, or mastering. A requested BPM, key, instrument, vocal character, or section transition is creative direction rather than a guarantee of perfect execution. Evaluate the actual file in its intended context.
Music 2.6 is also no longer MiniMax’s recommended generation. That does not make every existing 2.6 integration invalid, but it changes the default decision for new work. Keep 2.6 when reproducibility, an established evaluation, or a controlled migration justifies it. Test Music 3.0 when starting fresh or when output quality matters more than preserving a version-specific baseline.
For historical context, MiniMax Music 2.0 is now listed as a legacy model. Its guide should be used as an earlier-generation reference, not as a source of current Music 2.6 request fields.
Rights and Responsible-Use Checklist
API access and payment do not automatically answer copyright, voice, publicity, distribution, or commercial-use questions. MiniMax’s Music Creation Terms require users to own or hold the necessary rights and permissions for submitted content, including lyrics, music, recordings, and voices. They also require appropriate authorization for another person’s voice and place responsibility on the user for lawful input and output use.
- Submit original material or content for which the project has documented permission.
- Do not upload a person’s voice without the necessary authorization.
- Avoid prompts designed to impersonate a real person or reproduce protected songs, lyrics, or recordings.
- Review generated audio for unwanted similarity and suitability before release.
- Preserve prompts, lyrics, model IDs, settings, output records, consent, and license evidence.
- Follow applicable disclosure rules; MiniMax’s music terms state that users are responsible for clearly labeling AI-generated output to avoid misleading others.
- Obtain legal advice for a high-risk release, rights dispute, branded campaign, or use involving third-party music or voices.
This page is technical and editorial information, not legal advice, a license grant, or a promise that a particular output can be used commercially.
MiniMax Music 2.6 FAQ
Is MiniMax Music 2.6 still available?
Only for an eligible existing paying account. MiniMax closed new paid API access on August 20, 2026 and discontinued music-2.6-free. The schema still displays both IDs, so verify the account console before use.
What is the official Music 2.6 model ID?
An eligible existing paying account uses music-2.6; do not use music-2.6-free, which is discontinued. Namespace-style names used by another provider are not MiniMax’s official request strings.
How much does Music 2.6 cost?
MiniMax lists the paid model at $0.15 per up-to-5 minutes of music. The schema still shows music-2.6-free and a historical 3-RPM value, but the August 20 notice says the free API is discontinued. Pricing can change, so confirm the official pay-as-you-go page before production use.
Can Music 2.6 generate instrumental music?
Yes. Set is_instrumental to true and provide a prompt between 1 and 2,000 characters. Lyrics are not required in instrumental mode.
Can Music 2.6 generate lyrics?
Yes. With lyrics_optimizer set to true and the lyrics value empty, the API documents automatic lyrics based on the prompt. MiniMax also offers a separately priced Lyrics Generation/Editing service.
Does the Music 2.6 model create covers?
Cover was part of the Music 2.6 release, but the hosted reference-audio workflow uses music-cover only for an eligible existing paying account; music-cover-free is discontinued. Do not send cover-only fields to music-2.6.
Should a new project use Music 2.6 or Music 3.0?
Music 3.0 is MiniMax’s recommended text-to-music model and is the stronger default to evaluate for a new integration. Keep Music 2.6 when a measured comparison, compatibility requirement, or version-specific workflow supports that choice.
Does paying for the API guarantee commercial rights?
No such guarantee should be inferred from payment alone. Review MiniMax’s applicable terms, the rights in every input, consent for voices, the generated result, and the requirements of the intended platform or jurisdiction.
Official Sources and Verification Notes
- MiniMax Music Generation API reference — model IDs, authentication, endpoint, fields, character limits, audio options, outputs, and errors.
- MiniMax Music Generation guide — Music 3.0 highlights and current Cover workflows.
- MiniMax model catalog — Music 3.0, Music 2.6, Music Cover, and legacy categorization.
- MiniMax model release notes — dated release sequence.
- MiniMax pay-as-you-go pricing — legacy hosted prices and the current new-user availability notice.
- MiniMax rate limits — RPM and concurrent-connection table.
- MiniMax Music 2.6 launch article — April 10 release claims and Cover introduction.
- MiniMax Audio Music Creation Terms of Use — input rights, voice consent, prohibited uses, and AI-output notices.
Verification method: operational facts are prioritized from the live API schema, followed by the model catalog, pricing, rate-limit documentation, and dated MiniMax announcements. Promotional performance statements are attributed to MiniMax and kept separate from API guarantees. This site is independent and is not affiliated with or endorsed by MiniMax.