Rumble-2-Flash
Amsterdam startup Reson8's open-weights 0.5B speech-to-text model for 40+ languages, released under Apache 2.0 — it transcribes an hour of audio in about half a second and matches models 3-60x its size on accuracy.
Why it matters
A 0.5B model that scores 6.96% average WER on the Hugging Face Open ASR multilingual track (ahead of Whisper Large v3 and on par with Meta's 7B omniASR-LLM) and matches Nvidia's 30B Nemotron 3.5 Transcribe on a 302-hour real call-centre benchmark — near-enterprise accuracy that runs anywhere.
What you could build with it
A small team building a meeting notetaker or call-centre analytics tool could self-host this model and transcribe thousands of hours of calls for roughly €10 per thousand hours, with automatic language detection across 40+ languages and no per-minute cloud transcription bill.
Does it hold up?
Promising but early — vendor benchmarks are strong (6.96% multilingual WER, 7.86% on real call-centre audio) and it runs natively in Transformers/vLLM with no custom code, but no independent third-party deployments have been reported yet in the first day.
Built with Rumble-2-Flash
- Reson8 releases Rumble-2-Flash speech model as a free open-weights downloadsiliconangle.com · Launch coverage: open-weights release follows the Nov 3 API debut; models said to be in production use for meeting notetakers, AI receptionists and medical transcription.
- Reson8 releases open speech recognition model Rumble-2-Flashwattba.eu · Dutch coverage of the Nov 18 Hugging Face release, the Apache 2.0 licence, 40+ languages including Dutch, and the €0.01/hour hosted API.
- Reson8 open-sources 0.5B speech model for 40+ languagesoossa.com · Writeup noting native Transformers/vLLM support with no custom code, 7,000+ RTFx throughput, and the 17.6% vs 39.8% WER gap against Whisper Large v3 on low-resource languages.
Learn more
First spotted on hf: source.
More AI models
Edge0-145B: open 145B MoE that runs on a single consumer GPUEdge0 open-sourced 145B and 120B latent-MoE models derived from Qwen3.8-Flash-Next, plus an inference framework that…models · JEV 0.74Strata: run a 125B model on a gaming PCFree, MIT-licensed local inference engine that runs Qwen3.8-Flash-Next (125B MoE, about 6B active per token) on…models · JEV 0.73Qwen4-Max: Alibaba's new flagship model goes liveAlibaba's 2.4T-parameter flagship is now generally available via API, with a 1M-token context, native vision, and…models · JEV 0.67smolmaxxing: run 16B LLMs on a 1 GB TV box or a 2013 iPadAn MIT-licensed llama.cpp patch set that runs language models up to a 16B MoE on devices with about 1 GB of usable RAM…models · JEV 0.65Blue TTS v3: a 388MB open text-to-speech model with voice cloningAn open-source 388MB text-to-speech model (MIT) with voice cloning from about 10 seconds of speech, 7 languages, 15…models · JEV 0.65Haru-4BOpen full-duplex spoken-dialogue foundation model that listens and speaks simultaneously (7.5 Hz frames, 133 ms per…models · JEV 0.63
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.