
Rumble-2-Flash
Visit Rumble-2-Flash websiteOpen 0.5B multilingual speech recognition model with controllable output formatting for 45 languages
- Category
- Models, Audio & Voice
- Made by
- Reson8
- Pricing
- Open source
- Added
- 18 Nov 2026
- Our coverage
- Read our story
About Rumble-2-Flash
Rumble-2-Flash is a 0.5 billion parameter automatic speech recognition model built on Qwen3-ASR architecture. It supports 45 languages and introduces controllable output via special tokens for digits (words vs numerals), filler words (keep or drop), verbatim repeats (preserve or collapse), and forced language selection. Available under Apache 2.0 licence on Hugging Face with Transformers and vLLM integration, plus a Gradio demo. Benchmarks show competitive WER on Common Voice, MLS, VoxPopuli and Fleurs datasets.
Learn by doing
Relevant guides
Put Rumble-2-Flash to work with step-by-step guides from TechWhoKnows.