Reson8 has open-sourced Rumble-2-Flash, a 0.5B-parameter speech recognition model, under the Apache 2.0 license. This model is designed for over 40 languages and includes automatic language detection capabilities.
Rumble-2-Flash demonstrates high performance, processing an hour of audio in approximately half a second, achieving over 7,000 Real-Time Factor (RTFx) on full batches. It offers output control features such as numbers as digits or words, filler word handling, verbatim repeats, and forced language. The model is compatible with Hugging Face Transformers and vLLM, requiring no custom code.
On a 302-hour call-center benchmark, Rumble-2-Flash performed on par with NVIDIA's Nemotron 3.5 Transcribe (30B), despite being significantly smaller.
The model utilizes the Qwen3-ASR architecture for immediate compatibility with existing frameworks. Rumble-2-Flash was developed through on-policy distillation from Reson8's larger production model, Rumble-2. This distillation method involves the student model transcribing audio and the teacher model grading the transcript token by token, providing feedback on specific mistakes.
To address the challenge of differing tokenizers between teacher and student models, a technique called GOLD was employed. GOLD merges tokens where the two tokenizers split text differently, ensuring a consistent comparison basis for both models.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's briefNew every morning, and the back catalogue is archived by date.
Reson8 has open-sourced Rumble-2-Flash, a 0.5B-parameter speech recognition model under the Apache 2.0 license. This model supports over 40 languages, offers automatic language detection, and performs comparably to larger models while being significantly faster.