Rumble-2-Flash: an open 0.5B speech model for 40+ languages
Summary
- Reson8 has open-sourced Rumble-2-Flash, a 0.5B-parameter speech recognition model supporting over 40 languages under the Apache 2.0 license.
- The model utilizes the Qwen3-ASR architecture and was built using an on-policy distillation technique called GOLD to bridge different tokenizers.
- It natively supports Hugging Face Transformers and vLLM without requiring custom code.
- Rumble-2-Flash achieves competitive accuracy against significantly larger models on the Open ASR Leaderboard and call-center benchmarks.
Why it matters
Rumble-2-Flash offers high-speed, multi-lingual speech recognition in an extremely compact 0.5B parameter package, lowering deployment costs for developers while matching the performance of much larger models.