Content Deep Dive

Rumble-2-Flash: an open 0.5B speech model for 40+ languages

Blog post from Hugging Face

Post Details
Company
Date Published
Author
Reson8, Mark 't Hart, Eustache Le Bihan, and Eric Bezzam
Word Count
764
Company Posts That Month
60
Language
-
Hacker News Points
-
Post removed?
No
Summary

Reson8 has open-sourced Rumble-2-Flash under the Apache 2.0 license, a 0.5-billion-parameter speech recognition model supporting more than 40 languages, automatic language detection, and transcription options such as digit formatting and verbatim output. Built on the Qwen3-ASR architecture and distilled from the larger Rumble-2 model using a method called GOLD to handle differences between tokenizers, it runs natively with Hugging Face Transformers and vLLM. On the Open ASR Leaderboard, it records a 6.96% average error rate across nine languages and 6.6% English word error rate, while reaching more than 7,000 times real-time throughput on full batches. In a 302-hour benchmark of call-centre audio spanning 14 English accents, it achieves 7.86% word error rate, close to NVIDIA’s much larger Nemotron 3.5 Transcribe. The model weights are available on Hugging Face, and the model can also be accessed through Reson8’s API.

Trends Found in this Post
Trend Post Mentions Total Month Mentions Posts Companies MoM
LLM 1 3,355 742 191 -49%
MLX 1 11 5 3 -80%
Use This Data

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.