# Reson8 releases Rumble-2-Flash speech model as free open-weights download

published: 2026-11-18T20:05:55.869Z
topic: infrastructure
source: https://siliconangle.com/2026/11/18/reson8-releases-rumble-2-flash-speech-model-as-a-free-open-weights-download/
url: https://aiagentnews.io/stories/reson8-releases-rumble-2-flash-speech-model-as-free-open-weights-download

Reson8 has released Rumble-2-Flash, a 500 million-parameter transcription model covering more than 40 languages, as a free open-weights download under an Apache 2.0 license. The company says the model also performs well on lower-resource languages and is offered through its API from €0.01 per hour of audio.

Amsterdam-based Reson8 has released Rumble-2-Flash, a 500 million-parameter speech transcription model for prerecorded audio, as a free download with open weights. The model supports more than 40 languages, and the company is aiming it at developers and researchers who need recognition beyond English. Reson8 says its models already run in meeting notetakers, AI receptionists, customer support voice agents and medical transcription.

Reson8 reports a 7.09% average word error rate across seven languages in its own runs of the multilingual tests on Hugging Face's Open ASR Leaderboard, and about 5% in English. On a separate set of 10 lower-resource languages, it reports a mean word error rate of 17.6%, compared with 39.82% for OpenAI's Whisper Large V3 on the same test. The model debuted in Reson8's API on Nov. 3, while custom vocabulary of up to 50,000 terms remains limited to its flagship Rumble-2 model.

The model is available under an Apache 2.0 license, and the hosted API starts at €0.01 ($0.001) per hour of audio. Reson8 says its service runs on its own GPUs in the Netherlands and keeps no audio or transcripts, and it claims ISO 27001 certification along with GDPR and EU AI Act compliance. The company raised €5 million in pre-seed funding in March, led by Balderton Capital.

## Why it matters if you build agents

Developers building agents that handle non-English calls or recordings now have an open-weights speech model they can download, as well as a hosted API option. The accuracy figures are Reson8's own benchmark runs, so test the model on your target languages and audio before relying on them. Custom vocabulary is not available in the open model, so domain terms such as medication names may need a separate workaround.
