Amsterdam-based startup Resonate Labs B.V., known as Reson8, has released its Rumble-2-Flash speech transcription model with open weights. The 500 million-parameter model is designed for rapid processing of pre-recorded audio and supports over 40 languages.
The model is available for free under an Apache 2.0 license. Users can also access it via the company's API, with pricing starting at €0.01 per hour of audio. The service runs on the company's own GPU infrastructure in the Netherlands and maintains ISO 27001 certification.
In internal tests, Rumble-2-Flash achieved a 7.09% word error rate across seven languages and a 5% error rate for English. For 10 lower-resource languages, the model recorded a 17.6% error rate, compared to 39.82% for OpenAI's Whisper Large V3.
Reson8, founded by Raoul Ritter, Thomas Kluiters, and Jarno Verhagen, raised €5 million in a pre-seed funding round in March. The round was led by Balderton Capital with participation from NP-Hard Ventures.






