The GPTunneL platform has launched Grom Zvuk, a proprietary AI music model. The service operates on the company's own servers and allows users to generate tracks up to 5 minutes long, with the potential to extend this limit to 6 minutes. The model supports both vocal and instrumental compositions.
Grom Zvuk features tools for creating covers by applying custom lyrics to existing music, as well as changing genres and styles for established tracks. The output is provided in 48 kHz stereo. The architecture utilizes two experts under a shared attention mechanism to handle compositional logic and acoustic signal generation.
The model supports multiple languages, including Russian, English, Chinese, and Japanese, with a focus on correct prosody and stress for Russian text. In internal testing, Grom Zvuk achieved a text accuracy score of 85/100 and a sound quality score of 83/100, placing it among the top three models globally for text reproduction.
The base cost for generating a track is 8 ₽, with a 50% discount available until the end of September. GPTunneL also offers API access for business clients with scalable infrastructure. The platform currently serves 1.5 million users and provides access to over 100 AI tools, including its own Grom family of models.
