Updated

Rumble-2-Flash (open weights)

Reson8 open-sourced Rumble-2-Flash under Apache 2.0, making its 0.5B speech-recognition model available as downloadable weights. It supports automatic language detection across 40+ languages and runs with Transformers and vLLM.

By No Code MBA · Facts checked

Provider
Reson8
Announcement date
Status at last check
Generally available

What changed

This is the open-weights release of Reson8's Rumble-2-Flash, previously offered through its API. The released 0.5B speech-recognition model supports 40+ languages, automatic language detection, and controls for digits, fillers, and verbatim repeats. Reson8 reports strong speed and benchmark results, with throughput dependent on batch size.

Access and availability

Download the weights from the official Reson8/Rumble-2-Flash Hugging Face repository under Apache 2.0. Reson8 also says the model is available through its API.

Official access documentation

What you can build

Evaluate it on representative recordings in the languages and acoustic conditions your workflow handles, then review the transcripts before use.

Limitations

The model is for speech recognition, not a general-purpose model. Reson8 reports benchmark results and throughput; the latter depends on batch size and hardware.

Sources and release history