Artificial intelligence has already taught us how to properly compose a resume, write emails, draw cats in the style of Picasso, and solve complex math problems with childlike ease. All well and good. But OpenAI is now trying something that goes beyond Excel spreadsheets and aesthetically pleasing memes: using OpenAI Jukebox to create a song that people would actually want to hear more than once.
Although there is no official release date, the same question is circulating in the tech underground and research circles: Is it OpenAI preparing for a new version of Jukebox – OpenAI Jukebox 2.0? This is the experimental AI model that years ago proved it could compose a song, sing the lyrics, and even stylishly imitate Elton John, Nirvana, or anyone who has ever performed on MTV—all without a single human note. Well, at least not one that would come from an actual vocal cord.
Now, a few hints – in the form of academic publications, quietly updated GitHub pages, and “accidentally” liked tweets – suggest that a new version of this musical wonder might be happening. Maybe soon. Maybe at the end of April. Maybe not until the summer. Just like in the music industry – the release is “TBA”, but the audience is already waiting with headphones on.
OpenAi Jukebox: An AI that can do more than just repeat C major
When OpenAI first introduced Jukebox, critics rightly raised eyebrows. A model that generates music from scratch, including vocals, sounds like a futuristic scenario that a music producer with too much time and too few artists would have imagined. But Jukebox worked. By analyzing more than a million songs—including lyrics, genres, artists, and even the year of release—it created a new form of musical composition: one that’s based not on instruments, but on data.
The model didn't create notes, it created sound. And it did so directly in audio format, in studio-quality recording. You were listening to a song that never existed, but someone could easily mistake it for a lost Radiohead demo from 1998.
How is an artificial song born? Slowly, patiently, and with multiple layers of transformers
Technically, Jukebox works by first encoding raw audio into a compressed, symbolic language—an approximation of musical DNA. This encoded record is then processed by three layers of neural networks, each contributing in its own way to melody, rhythm, vocals, and sonic texture. The final step is reverse decoding—where the AI literally folds the sound wave back together to create something that humans can hear, but the AI merely “predicts.”
The whole thing works like a kind of digital studio, where the producer never needs a break, the singer never stutters, and the author's inspiration is not dependent on the weather. The only problem: generating a song in its current form takes almost nine hours for a single minute of sound. Which means that Jukebox is now more of a slow-motion symphony composer than an instant hitmaker.
What do the rumors about the new version suggest?
In recent weeks, there has been increased activity among developers associated with OpenAI's audio division. Commit logs have appeared on GitHub mentioning model optimizations. Several researchers have hinted in podcasts (and even more informally in X/Twitter conversations) that work is underway. new architecture, which is supposed to enable faster generation, better vocal articulation, support for multiple languages, and – what is especially interesting – better understanding of song structure.
This means that the new version could, for the first time, generate songs with clearly defined choruses, intros, bridges, and endings – something you hear in a real song, not a digital sketch.
Add to this the rumors of a more accessible user interface and integration with other creative tools (such as APIs for content creators), and it becomes clear: something is cooking, and it's not far away.
AI in music: assistant or competition?
The new version of Jukebox will reopen the classic question: does artificial intelligence replace creativity or augment it? Critics will say it's a technological decontextualization of art - a song without a soul, an emotion without an experience. But advocates will be thrilled, because Jukebox offers something completely new: the possibility of creating music as an idea, not as an execution.
Imagine an independent artist who creates a professional-sounding song in the style of Massive Attack without a studio budget. Or an indie developer who generates a soundtrack for their video project with a few lines of text. Or a student who writes a song about their crush and listens to it in the style of Arctic Monkeys.
This is no longer science fiction – this is Jukebox. Or to be more precise: Jukebox, a version that doesn't exist yet. But we all know it's coming.
Conclusion: Sometimes the best sound comes from the silence before the premiere
Nothing has been officially confirmed yet. OpenAI is silent, but its silence is suspiciously loud. The atmosphere among researchers is similar to that of Kanye West albums – no one knows when it will be released, but everyone is already writing reviews. If Jukebox 2.0 is what the rumors predict, we may soon hear songs that are no longer based on the inspiration of a single individual, but on the collective memory of millions of songs. And this with a voice we have never heard before, but that sounds surprisingly familiar.
In other words: if you feel like there's nothing new in music these days – just be patient a little longer. Maybe the next hit is already being generated. Quietly, slowly, in a cloud. And when it's finished, no one will be singing it – but you'll still be singing along.
