Recensione di Mureka AI (2026): test musicali di Mureka V9, prezzi e API

Our Mureka AI review found a music generator with fast full-song output, consistent technical delivery, and a serious duration-control problem. All eight new Mureka V9 tasks succeeded in the test environment and returned 44.1 kHz stereo MP3 files. The shortest generation finished in 57.62 seconds. However, every one of the six requests that asked for 90 seconds overshot the target, sometimes by more than two and a half minutes.

Il verdetto in breve: Mureka V9 is useful when you want a complete musical interpretation and can edit or select between takes. It is a weak fit when the final duration must match the prompt. There is also an important version update: Mureka’s consumer model endpoint now labels V9.5 as the current and default model. The hands-on evidence in this review is for V9, not V9.5.

This review combines eight newly logged V9 generations with three older listening baselines. The new set supports objective conclusions about output length, generation time, file format, loudness, peak level, stereo presentation, and repeatability. Human review of lyric accuracy, Chinese pronunciation, section order, and voice leakage is still pending for those eight files, so those judgments are not presented as finished results.

Mureka V9.5 is now current, but these music tests use V9

As of August 11, 2026, Mureka’s official consumer model response identifies mureka-9.5 as both current_model e default_model. The same response still lists mureka-9 as an available model. On the developer side, Mureka advertises V9.5 as early access with the positioning “More Natural · More Expressive · Truer to Your Prompt.”

That distinction matters. This article does not quietly relabel V9 outputs as V9.5 results. The eight new tasks used the test routes mureka-v9-song e mureka-v9-bgm, and the three listening baselines also used V9. Treat the findings as a V9 reliability record and a useful benchmark for judging whether V9.5 improves the same weak points.

Mureka’s public API changelog has not yet added a V9.5 release row. Its latest named model entry remains the enhanced `mureka-9` release dated April 9, 2026. The consumer model response and API early-access banner are therefore the fresher sources for V9.5 availability, while the changelog remains the clearest dated source for the V9 update tested here.

Mureka announced V9 through its official March 27, 2026 launch post. That date is historical context for the tested model, not evidence that V9 remains the newest consumer option.

Mureka AI review: quick verdict

Yes, with one major qualification. Mureka can turn a detailed brief into a full-length song or background track in about a minute of processing time, and our successful files were technically consistent. What it did not do reliably was honor a requested 90-second duration.

For a direct category comparison, our earlier ElevenLabs vs Lyria 3 Pro vs Mureka: test musicale used the same three briefs across all models. Mureka produced the longest and most stylistically independent outputs in that set, but also created the greatest editing burden.

What is Mureka AI?

Mureka AI is an AI music-generation platform for complete songs, instrumentals, lyrics, soundtrack work, and production tools. The consumer product separates song creation from background-music workflows, while the developer platform exposes generation, query, remix, extension, stem, transcription, and soundtrack operations for automated applications.

The product is broader than a prompt box. Paid consumer plans include reference-track workflows, remixing, editing, lyrics-video export, and commercial rights for newly made songs. Premier adds Mureka Studio, up to 12-stem splitting, WAV, MIDI, and DAW-session exports, custom singer workflows, and professional voice cloning.

That breadth also creates a decision point. If you need one prompt to combine dialogue, ambience, Foley, and music, our Test di Seed Audio 1.0: voci, effetti sonori e musica cover a more unified scene-generation approach. Mureka is the more music-centered product.

How we tested Mureka V9

The new hands-on set ran through the authorized GLBGPT/Broly Anywhere Tasks test environment with WaveSpeed upstream, not the official Mureka consumer app and not a native call to api.mureka.ai. We used two exposed V9 routes: mureka-v9-song for vocal music and mureka-v9-bgm for instrumental background music.

We designed three tasks:

1. Three repeats of a 90-second contemporary synth-pop song with fixed lyrics.
2. Three repeats of a 90-second cinematic BGM brief that prohibited vocals and voice-like textures.
3. Two repeats of a city-pop ballad with exact English and Chinese lyrics.

Every successful task reported 75,000 quota units, cost_usd: 0.045, e margin_usd: 0.105. Across eight tasks, that was 600,000 reported quota units. Those fields belong to the Broly/WaveSpeed route; they are non Mureka’s consumer subscription price, Mureka Gold, or a bill from the official Mureka API.

The test environment initially returned an insufficient-quota error before the successful run set began. Those rejected submissions did not create tasks, return audio, or report a task charge. Once access recovered, the eight logged generation tasks completed successfully.

Eight new Mureka V9 music tests

Test 1: lyric-song repeatability

The first prompt requested a 90-second synth-pop song at 112 BPM in A minor with a warm alto lead, exact lyrics, a defined arrangement, and a clean final chord. We submitted the identical prompt three times.

The mean output was 165.12 seconds—75.12 seconds longer than requested. Generation time was much more stable than output length, staying within a 5.03-second range. That is useful for queue planning, but it does not solve the editing problem created by a song that is almost twice the requested length.

Play the three raw outputs below.

Test 2: no-voice cinematic BGM repeatability

The second task asked for a 90-second science-fiction exploration cue at 82 BPM in D minor. The prompt prohibited vocals, speech, whispers, choir, and human voice-like sounds, while specifying felt piano, low strings, an analog pulse, restrained percussion, late brass, and a resolved ending.

The mean output was 207.19 seconds, or 117.19 seconds over the target. The 51.84-second gap between the shortest and longest repeats is the clearest repeatability warning in the new data. P02-R01 also reached -0.18 dBTP after four-times oversampling. It did not cross 0 dBTP, but it left the least peak headroom of all eight files.

Play the three raw outputs below.

This objective pass cannot tell us whether the new BGM files contain a vocal-like “ah,” whether the brass entered in the final third, or whether the piano motif remained recognizable. Those are listening questions. If your real goal is synchronized video with generated sound, the separate Veo 3.1 sound-generation guide explains when a video-native model may reduce the handoff between picture and audio.

Test 3: English and Chinese lyrics

The third task used exact English and Chinese lyrics in one city-pop ballad. No duration was specified, so this test measures variation rather than target accuracy.

The same request produced outputs 15.40 seconds apart. P03-R02 also contained 3.35 seconds at the end below -50 dBFS. That may be harmless tail space or an editing nuisance, but a human listener still needs to decide whether the musical ending itself feels complete.

Play the two raw outputs below.

We have not published a new lyric-fidelity or pronunciation score for these two clips. If exact words matter, transcribe the complete vocal and compare it line by line; our overview of ChatGPT audio transcription options covers practical ways to start that QA, but native-language listening remains necessary for pronunciation.

Technical output quality

Mureka’s V9 test-route delivery was consistent even when the creative duration was not. All eight outputs decoded as 44.1 kHz stereo MP3 files. The song route delivered approximately 192 kbps, while the BGM route delivered approximately 256 kbps.

No decoded samples exceeded 0 dBFS, and no samples reached the -0.01 dBFS near-clipping threshold. Integrated loudness ranged from -15.95 to -13.91 LUFS. That does not prove the mixes are musically balanced, but it does show that the files were delivered at usable levels without obvious sample clipping.

Stereo width changed between repeats. The BGM side-to-mid ratios ranged from -6.36 to -3.68 dB, while the lyric-song set ranged from -12.09 to -9.48 dB. In practical terms, identical prompts did not lock the same stereo presentation. Compare candidate takes in mono and on headphones before choosing one for a final mix.

Three listening baselines: where V9 sounded strong and where it drifted

The following observations come from the older three-model comparison, not from the unfinished listening pass on the eight new files. That review used editor notes cross-checked with Gemini 3.6 Flash audio analysis; it was not a human listening panel or a human MOS study.

Vocal synth-pop: distinctive interpretation, weaker exact control

The V9 song ran for 170.7 seconds from a 90-second brief. It developed a distinctive electronic, nu-disco-like identity and a long instrumental outro. A blind transcript flagged a likely lyric substitution, so the result felt more like an interpretation than a deterministic rendering of the supplied text.

That kind of vocal can still be useful for ideation. If the synthetic delivery feels too stiff, the techniques in our guide to making AI voice and writing sound more human can help you rethink phrasing before regeneration, although the final performance still depends on the music model.

Cinematic BGM: compelling composition, failed no-voice constraint

The V9 BGM baseline lasted 183.9 seconds. It created a mournful violin-led opening and built toward a later climax, but a voice-like “ah” texture broke the explicit no-vocals and no-human-voice-like-sounds instruction. Several requested elements were replaced by the model’s own orchestral interpretation.

This is not just a stylistic disagreement. A human-like texture can create extra cleanup when the music must sit under dialogue. For projects where characters also need speech, the Veo 3.1 Flusso di lavoro relativo a dialoghi, audio e sincronizzazione labiale shows why voice, music, and picture should be reviewed as one timing system.

Fixed structure: polished groove, softer section contrast

The eight-part alternative-pop baseline lasted 167.1 seconds. It had a polished pop-funk groove, a clear vocal hook, and a long guitar outro. The bridge and outro were less distinct than requested, and the final structure created more trimming work than the competing outputs.

What Mureka V9 does well

Fast full-song generation

The eight successful tasks completed in 57.62 to 86.17 seconds. For outputs lasting roughly 2.6 to 4 minutes, that is a productive ideation speed. Mureka can give you a complete arrangement before you would finish manually sketching the same sections in a DAW.

Strong willingness to interpret a brief

The older baselines did not sound like timid copies of the prompt. V9 introduced distinctive genre character, altered instrumentation, and expanded endings. That independence is valuable when you want the model to surprise you; it becomes a weakness when you need literal compliance.

Consistent delivery format

Every new file arrived as a playable stereo MP3 at the same 44.1 kHz sample rate. Song and BGM routes also showed consistent bitrate patterns. That makes automated download, cataloging, and preview workflows easier to design.

Separate routes for songs and BGM

The test environment exposed dedicated song and BGM routes, while Mureka’s official API includes lyrics-to-song, prompt-to-song, instrumental, and soundtrack operations. Developers can choose a workflow that matches the job instead of pushing every task through one generic endpoint.

Mureka V9 limitations

Duration instructions were not reliable

This is the strongest conclusion in the review. All three lyric-song repeats and all three BGM repeats missed the requested 90 seconds. The song group averaged 165.12 seconds; the BGM group averaged 207.19 seconds.

If a client needs exactly 15, 30, 60, or 90 seconds, budget for editing. Do not assume that placing the duration in the creative prompt will produce a broadcast-ready runtime.

Identical prompts produced different lengths and stereo behavior

The BGM repeats differed by 51.84 seconds, and the bilingual song repeats differed by 15.40 seconds. Stereo width also shifted across identical requests. Mureka behaves like a multi-take creative generator, not a deterministic renderer.

Negative constraints need listening QA

The older BGM baseline added a human-like vocal texture despite explicit prohibitions. The eight new BGM files require human listening before we can say whether that failure repeated. For now, “no vocals” should be treated as an instruction to verify, not a guarantee.

Voice cloning creates an extra privacy obligation

Premier promotes professional voice cloning and the ability to use a custom voice as an AI singer. Use only voices you own or have permission to process. The wider face-to-voice privacy risk discussion is relevant here: access to a feature does not replace consent, disclosure, or rights management.

Exact Mureka V9 prompts used in the new tests

These prompts are useful as starting templates, but the results above show that detailed wording does not guarantee exact length or repeatability.

Prompt 1: contemporary synth-pop with exact lyrics

Create one completely original 90-second contemporary synth-pop song with a warm, expressive alto lead vocal. Use 112 BPM and A minor. The mood is hopeful, intimate, and cinematic. Begin with muted electric piano and a soft analog-synth pulse. Add round bass and tight electronic drums in the first verse, build tension in the pre-chorus, and open into a wide, memorable chorus with restrained background harmonies. Keep the lead vocal clear and centered with clean instrument separation. No spoken introduction, rap, crowd sounds, imitation of a real artist, or fade-out. End on a clear final chord.

Perform the exact lyrics below. Do not add, remove, replace, or reorder words or sections.

[Verse 1]
City lights fade behind our backs
Silent streets now lead the way
Midnight whispers, no looking back
A new horizon breaks the gray

[Pre-Chorus]
Hand in hand, we step unknown
Chasing dreams we call our own

[Chorus]
We're alive in the quiet night
Finding stars beyond the skyline
Lost in shadows, yet so bright
Together, we'll redefine

[Verse 2]
Footsteps soft on empty roads
Every breath a story told
No promises, just endless skies
In your eyes, the future lies

[Final Chorus]
We're alive in the quiet night
Finding stars beyond the skyline
Lost in shadows, yet so bright
Together, we'll redefine

Prompt 2: cinematic BGM with no voice-like textures

Crea un brano strumentale cinematografico di 90 secondi completamente originale per una scena drammatica di esplorazione fantascientifica. Utilizza 82 BPM e la tonalità di Re minore. Non sono ammessi parti vocali, dialoghi, sussurri, cori o suoni simili alla voce umana. Inizia in modo sommesso con un motivo essenziale al pianoforte e archi gravi sostenuti. Introduci un sottile impulso analogico dopo l’apertura, poi costruisci gradualmente la traccia con archi sovrapposti e percussioni gravi contenute. Aggiungi gli ottoni solo nell’ultimo terzo. Raggiungi un chiaro climax emotivo verso la fine senza volume eccessivo o distorsione. Mantieni il motivo pianistico riconoscibile e preserva una chiara separazione tra pianoforte, archi, percussioni, pulsazione sintetica e ottoni. Utilizza timbri orchestrali realistici con un mix spazioso ma controllato. Evita gli effetti tipici dei trailer nella prima metà. Non imitare una colonna sonora, un compositore o un franchise esistente. Concludi con un accordo sostenuto e risolto piuttosto che con una dissolvenza in fondo.

Prompt 3: English and Chinese exact lyrics

Create one completely original city-pop ballad at 96 BPM with a warm alto lead vocal, clean bass, electric piano, restrained guitar, and a clear final chord. Keep the verses intimate and make the chorus wider without adding spoken words, rap, crowd noise, or an artist imitation. Perform the exact lyrics below. Do not add, remove, replace, translate, or reorder words or sections.

[Verse 1]
Neon rain is falling slow
Last train lights begin to glow

[Chorus]
Stay with me until the morning
Hold the line while night is turning

[Verse 2]
霓虹落在安静街口
末班列车带走等候

[Final Chorus]
陪我等到城市苏醒
别让回声消失无影

Mureka AI pricing in 2026

Mureka has separate consumer subscriptions and developer API billing. Do not compare a monthly Gold allowance directly with a per-call API price, and do not treat the Broly test-route fields as either one.

The consumer prices below come from Mureka’s official subscription product response and plan page, checked August 11, 2026.

Mureka’s live pricing data contains inconsistent daily-Gold copy for the Free plan, so this review does not publish a fixed daily number. Check the live Free card before relying on a daily quota. The paid-plan prices, 5,000/20,000 Gold allowances, and stated song/speech ceilings are consistent in the current plan data.

Which Mureka plan should you choose?

Scegliere Gratuito to test the workflow and V7.5-all without commercial use. Choose Pro if you mainly need song generation, MP3 downloads, references, remixing, and commercial rights. Choose Premier when production exports, stems, Studio, MIDI, DAW sessions, custom singers, or voice cloning are central to your workflow.

The “up to” song counts assume a particular Gold cost per generation. Real usage can change with the selected model, feature, or number of outputs, so treat the allowance as a plan ceiling rather than a promise that every project will reach it.

Mureka API pricing

Mureka’s official API checkout product feed returns a $10 prepaid entry product with one concurrent request, while the visible Tabella dei prezzi delle API shows $30 as the one-concurrency recharge amount. Both were live on August 11, 2026. Because these two official surfaces disagree, confirm the amount shown at checkout before paying. Neither figure is a monthly subscription.

L'attuale Mureka API pricing page lists the V9, prompt-to-song, BGM, and soundtrack rates below. Mureka’s developer-platform overview separately advertises the time-limited V9.5 early-access rate.

The V9.5 beta price is time-sensitive. Recheck the official platform after August 28, 2026 rather than assuming the $0.045 early-access rate will continue.

How the Mureka API works

Mureka’s Guida rapida ufficiale usi https://api.mureka.ai as the server and Bearer API-key authentication. The basic lifecycle is asynchronous:

1. Create an API key in the Mureka API dashboard.
2. Send a POST request to a generation endpoint such as /v1/song/generate.
3. Save the returned task ID and trace_id.
4. Poll GET /v1/song/query/{task_id} until the task reaches a terminal status.
5. Save the final media URLs and log trace_id for troubleshooting.

The documentation’s first request uses model: "auto", which is the safest default when you want Mureka to choose the available model for your account:

curl https://api.mureka.ai/v1/song/generate \
  -H "Authorization: Bearer $MUREKA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "lyrics": "[Verse]\nCity lights fade behind our backs...",
    "model": "auto",
    "prompt": "contemporary synth-pop, 112 BPM, A minor, warm alto vocal"
  }'

Never place the API key in browser-side JavaScript, public screenshots, article HTML, or a repository. This review verified the official documentation and pricing surfaces, but it did not run a native paid call against api.mureka.ai. The hands-on generations came from the separate Broly/WaveSpeed test route described above.

Commercial rights and production use

Mureka’s current consumer plan response states that Pro and Premier include commercial-use rights for new songs made under those plans. The Free plan is explicitly marked as non-commercial. Download and production options also differ: Pro centers on MP3, while Premier adds WAV, stems, MIDI, and DAW-session exports.

Commercial permission does not mean every input is risk-free. Do not imitate a living artist, upload music you do not control, or clone a voice without consent. Keep the prompt, account plan, source files, generation date, and license or ownership certificate with the project archive.

Mureka alternatives

Scegliere Mureka when you want fast, full-length musical interpretation and can trim or select between outputs. Its separate song and BGM routes are useful, and the official API pricing is easy to understand.

Scegliere ElevenLabs Music v2 when exact lyrics and recognizable section structure matter more. In our three-model baseline, ElevenLabs preserved the supplied words and eight-part structure more consistently than Mureka.

Scegliere Google Lyria 3 Pro when a cinematic brief, motif development, late orchestration changes, and a resolved ending are the priority. In the same comparison, Lyria followed the cinematic prompt more closely.

Scegliere Seed Audio 1.0 when the job combines dialogue, ambience, effects, and music inside one generated scene. It is a broader audio-scene model rather than a dedicated full-song platform.

If you want one place to compare AI audio workflows before committing to separate subscriptions, the tracked GlobalGPT audio generator workspace is the practical route to open next. That link uses a campaign parameter; it does not change the review’s test results.

Mureka AI pros and cons

Pro

Contro

Who should use Mureka AI?

Mureka is a strong fit for songwriters, video creators, game teams, podcasters, and developers who want a fast musical draft and already expect a selection or editing stage. It is particularly attractive when a long output is useful and an exact runtime is not.

It is less suitable for ad delivery, broadcast slots, tightly timed picture edits, or lyric-critical releases that must succeed in one unattended request. Those workflows need trimming, full-song transcription, native-language listening, and often a DAW.

Domande frequenti

Is Mureka V9 still the newest model?

No. As of August 11, 2026, Mureka’s consumer model endpoint identifies V9.5 as the current and default model. The official API platform also advertises V9.5 early access. This review’s hands-on files were generated with V9 routes, so the test results should not be presented as V9.5 performance.

Is Mureka AI free?

Mureka has a Free consumer plan with access to V7.5-all, a shared queue, and no commercial use. Its live pricing data contains inconsistent daily-Gold amounts for that plan, so check the visible Free card before planning repeated generations.

How much does Mureka AI cost?

The consumer Pro plan costs $10 per month or $96 per year and includes 5,000 Gold per month. Premier costs $30 per month or $288 per year and includes 20,000 Gold per month. Annual billing is about 20% cheaper than paying monthly for 12 months.

How much does the Mureka API cost?

The official checkout product feed shows a $10 prepaid entry product with one concurrent request and 12-month validity, while the visible official pricing table shows $30 for a one-concurrency recharge. Confirm the amount shown at checkout before paying. V9 lyrics-to-song and V9 BGM are listed at $0.045 per output, prompt-to-song at $0.30, and V9 soundtrack generation at $0.10. V9.5 early access is $0.045 per song until August 28, 2026, with a displayed standard price of $0.15.

Can Mureka generate instrumental music?

Yes. Mureka offers BGM and instrumental workflows, and its official API also includes soundtrack generation from an image or video. In our tests, the V9 BGM route returned three long stereo MP3 files, although all three exceeded the requested 90-second duration.

Does Mureka follow exact lyrics?

It can accept exact lyrics, but our older vocal baseline flagged a likely substitution, and the eight new files have not completed human lyric review. Generate more than one take, transcribe the full vocal, compare every repeated chorus, and listen for added words before publishing.

Can Mureka sing in Chinese?

Mureka accepts Chinese lyrics, and we generated two English-Chinese city-pop outputs with V9. A native human review of word accuracy, pronunciation, code-switching, and vocal identity is still pending, so this article does not assign a bilingual quality score.

Can I use Mureka songs commercially?

Mureka’s current consumer plan response grants commercial-use rights for new songs made on Pro and Premier. The Free plan is marked non-commercial. You still need rights to uploaded references and cloned voices, and you should preserve the relevant license or ownership documentation.

How long does Mureka take to generate a song?

The eight successful V9 tasks completed in 57.62 to 86.17 seconds through the Broly/WaveSpeed test route. That is not an official Mureka API latency guarantee. Queue load, route, model, requested output count, and platform can all change the time.

Does Mureka obey requested song duration?

Not reliably in our V9 tests. All three 90-second lyric-song requests returned 158.83 to 171.79 seconds, while the three 90-second BGM requests returned 189.47 to 241.31 seconds. Plan to trim the output when runtime matters.

Can I use Mureka on GlobalGPT?

This review does not claim a verified public GlobalGPT Mureka product route. The hands-on generations used the Broly Anywhere Tasks test environment with WaveSpeed upstream. That testing access is not the same as a public GlobalGPT product page, account plan, or consumer workflow.

Il verdetto finale

Mureka V9 is fast, technically consistent, and willing to turn a detailed prompt into a complete musical interpretation. The eight new tests also reveal its clearest production weakness: duration control. A 90-second request can become a song or BGM track lasting well over three minutes, and identical prompts can return meaningfully different lengths and stereo presentations.

That tradeoff is acceptable for ideation, songwriting, game concepts, and video projects with room to edit. Generate several takes, compare the full files, verify every lyric and negative constraint, then trim the best result. It is much harder to recommend for one-shot delivery into a fixed runtime.

The newest question is now V9.5. Mureka has made it the consumer current/default model and opened API early access, but this Mureka AI review does not pretend that V9 evidence answers V9.5 quality. The next meaningful test is a controlled V9-versus-V9.5 rerun using the same three prompts, with human listening focused on lyric fidelity, voice leakage, arrangement restraint, and ending control.