Understand this story
The English summary is being updated. See the reporting timeline below.
Latest development · Report
Reson8 Open-Sources Rumble-2-Flash Speech ModelDevelopments
1 confirmed developments
- Source:Hugging Face
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Nov 18
SiliconANGLE
Reson8 Open-Sources Rumble-2-Flash Speech Model
Reson8 has released the open-weight speech recognition model Rumble-2-Flash, supporting 40+ languages. The 500M parameter model achieves an average word error rate of 7.09% in multilingual tests, outperforming Whisper Large V3 on low-resource languages. The model is freely available under the Apache 2.0 license, with API calls starting at €0.01/hour.
Hugging FaceFirst-partySelected
Rumble-2-Flash Released: An Open-Source 0.5B Speech Model Supporting 40+ Languages
Rumble-2-Flash is an open-source 0.5B speech model supporting 40+ languages. Built on the Qwen3-ASR architecture, it can run with Transformers and vLLM. On the Open ASR Leaderboard, it achieves an average error rate of 6.96% on the multilingual track and an average WER of 6.6% on the English track, while throughput exceeds 7,000 RTFx with sufficiently large batches. The model weights have been released on the Hugging Face Hub at Reson8/Rumble-2-Flash, and the model is also available via API.
AI Trends