Suno drops V3 Alpha with 2-minute clips and pure instrumentals

Making full-length songs with generative AI used to feel like stitching together disjointed thirty-second audio snippets.

Suno just rolled out V3 Alpha access to its paying Pro and Premier members. To encourage testing and collect immediate feedback, the team dropped 300 free credits into subscribers' accounts.

The update brings several heavily requested improvements that make audio creation substantially more cohesive.

Two-minute clips and backward compatibility

The biggest leap in V3 Alpha is track length. The model can now generate continuous clips of up to two minutes in a single run, which is long enough to flesh out an entire verse-chorus section.

Generation speeds are noticeably faster, alongside upgrades to general audio fidelity and vocal expressiveness. Suno also added a dedicated instrumental toggle and broadened its foreign language support.

There is also a new feature called Continue from anywhere. Crucially, it works retroactively, meaning you can pull up an unfinished track generated with V2 and build out the rest inside V3 Alpha.

Subscribers can flip back to V2 at any point from the web app dropdown if they prefer the older model.

Longer generations come with weird hallucinations

Suno was surprisingly upfront about where V3 Alpha still falls short in its release notes.

The team noted that the model still struggles to honor strict technical constraints like BPM and key signatures. Final mixes and mastering are also far from polished studio quality.

More importantly, longer clip limits open the door to audio hallucinations. Suno warned that feeding short prompts into the new model can cause it to spiral into bizarre audio artifacts or nonsensical vocal riffs.

Prompting habits from V2 will not necessarily translate cleanly, and the company expects to tweak model parameters without warning as testing progresses.

Why it matters

For content creators and video editors, two-minute coherent generations and a dedicated instrumental mode remove a huge roadblock. You no longer have to crossfade short segments just to get decent background music.

Still, if you are expecting radio-ready productions, it is not worth rushing to buy a subscription right now. The missing control over exact tempo and key means professional producers will still hit walls quickly.

We think Suno's rollout strategy is smart. Instead of keeping a flawed model behind closed doors until it is bulletproof, they are leveraging their most invested users with free credits to crowdsource the necessary alignment data.

The trade-off is clear: as generative music models grow longer memory spans, reigning in their hallucinations without killing melodic flair will be the defining battle for the space.