Hugging Face has released Rumble-2-Flash, a new open-source speech recognition model with 0.5 billion parameters. This model supports over 40 languages and demonstrates competitive performance against much larger models, including NVIDIA's Nemotron 3.5 Transcribe. Rumble-2-Flash achieves high accuracy and speed, processing an hour of audio in approximately half a second, and is designed to run efficiently within the Hugging Face Transformers and vLLM frameworks. AI
IMPACT This release provides a highly efficient and multilingual speech recognition model, potentially lowering the barrier for integrating advanced ASR capabilities into various applications.
RANK_REASON Open-source release of a new speech model by a major AI lab (Hugging Face). [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Apache Software License 2.0
- Hugging Face
- Meta
- Nemotron 3.5 Transcribe
- Nvidia
- omniASR-LLM 7B
- Qwen3 ASR
- Rumble-2-Flash
- transformers
- vLLM
- Whisper
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →