Pianissimo

Pianissimo is an open-weight speech-recognition model for Swedish, from Klang AI, a Swedish company, released on 23 September 2026. It is a fine-tune of NVIDIA's Parakeet v3 architecture with 0.6 billion parameters, trained on about 50,000 hours of Swedish speech. It takes audio and returns text.

What Pianissimo offers

• Swedish transcription: converts Swedish audio to text.
• Error rates: 6.51 per cent on FLEURS Swedish, 4.46 on Common Voice and 4.85 on Klang's own dialect test set, as the developer reports.
• Speed claim: the developer says it matches KB-Whisper large while running 64 times faster.
• Access: open weights, and a hosted route through Berget AI's API.

Who it works with

The model suits Nordic organisations that transcribe meetings, calls or archives in Swedish and want to run it on their own hardware.

Key facts

• Licence: CC BY 4.0, which permits commercial use with attribution.
• Model ID: pianissimo-sv.
• Speed figures: 2,500 times real time on one GPU, 3,600 on the product page and more than 4,000 elsewhere, which disagree with each other.