Reson8/Rumble-2-Flash: Open 0.5B speech recognition model for 40+ languages
Reson8 has open sourced Rumble-2-Flash under the Apache 2.0 license, a 0.5B-parameter speech recognition model supporting over 40 languages. The model uses the Qwen3-ASR architecture and was trained via on-policy distillation using a technique called GOLD to bridge tokenizer differences. It achieves competitive accuracy and high processing speeds, handling an hour of audio in about half a second. Open article →