Reson8 Releases Free Open Transcription Model

Rumble-2-Flash, a 500 million-parameter model supporting more than 40 languages, is now available as an open-weights download.

Updated on Nov. 18, 2026 in Language Learning

Bold flat-color editorial illustration of a row of matte rectilinear blocks in varying heights and muted colors rising in sequence, evoking multilingual transcription access.
Reson8 released its Rumble-2-Flash transcription model as a free open-weights download supporting more than 40 languages, after first offering it through its API on Nov. 3. AI Illustration. Upload story photo >

Amsterdam-based Resonate Labs B.V. released its Rumble-2-Flash transcription model as a free open-weights download under an Apache 2.0 license. The company, operating under the Reson8 name, says the model supports more than 40 languages.

Why it matters

Reson8 says leading speech models are trained primarily on English, though only about one in 20 people speak it as a first language. An openly available multilingual model could broaden access to transcription tools for speakers of other languages.

Rumble-2-Flash is a 500 million-parameter model with a claimed average English word error rate of about 5%, and it is priced from €0.01 per hour of audio through Reson8's API. The company's error-rate comparisons come from its own tests, and independent verification has not been reported.

The players

Reson8

Reson8 is the name under which Resonate Labs B.V., an Amsterdam-based company, offers its speech transcription services and models.

The details

Rumble-2-Flash was first offered in Reson8's API on Nov. 3 and was released as a free open-weights download on Nov. 18. Reson8 says it hosts the service on company-owned GPUs in the Netherlands, retains no audio or transcripts, and claims ISO 27001 certification.

Timeline

  1. • Nov. 3: Rumble-2-Flash debuted in Reson8's API.

  2. • Nov. 18: Rumble-2-Flash was released as a free open-weights download.

Language Landscape

Reson8 positions Rumble-2-Flash against OpenAI's Whisper Large V3, which it reports had a 39.82% mean error rate on 10 lower-resource languages in the same test. The release places a free, openly licensed alternative into a market where open speech-recognition models have so far been dominated by English-trained systems.

Because the model is released under an Apache 2.0 license, developers and learners can download and run it without paying per-use fees. Users who prefer not to self-host can access it through Reson8's API at prices starting from €0.01 per hour of audio.

The takeaway

Free, openly licensed speech tools can make transcription more usable for speakers of many languages, not only English. Language learners and educators may find such models useful for practicing listening and reviewing spoken material.

Further reading

Explore more on Language Learning to see how speech tools are reshaping how people study new languages.

Source note: This article includes information reported by SiliconANGLE.