Rumble-2-Flash. Open weights.

0.5B parameters · Apache 2.0

Download
Contact sales
Skip film
Scroll to play the film
Pause film
00:00 / 00:54
Sound off
Every voice deserves to be understood. Now the model is yours.
Download
Contact sales for Enterprise
Scroll down to read about open weights
Pause film

Everything you need to know before download

Parameters
0.5B parameters
One of the smallest models on the Hugging Face Open ASR Leaderboard
Licence
Apache 2.0
Download it, run it, fine-tune it. It’s yours.
Languages
40+
Available on
Hugging FaceDownload now.
WER
6.96%
Avg. WER across 9 languages, Open ASR (HF)

Small model, big ears

We wanted to see how small a model could be without the accuracy falling off. The target was a model small enough to run on everything, from voice notes to full call archives, while still holding up on complex, noisy audio.

The main challenge: real phone calls

Transcribing clean audio is easy, phone calls are not. The audio is squashed and noisy, with fast-talking people on opposite sides of the globe. So we ran rumble-2-flash on AppTek’s call-centre test, scored with AppTek’s own script.

It got about 92% of the words right. That’s on par with NVIDIA’s newest speech model, which is 60 times bigger, and ahead of Mistral, IBM, Cartesia, Deepgram, ElevenLabs and AssemblyAI. One model does better: Alibaba’s Qwen3-ASR 1.7B, more than three times our size.

Real phone calls, 14 English accents
Words missed in every 100. Shorter is better. AppTek call-centre test, 302 hours.
Qwen3-ASR 1.7BAlibaba
7.4
Nemotron 3.5 Transcribe (30B)NVIDIA
7.78
Rumble-2-Flash Open (0.5B)Reson8
7.86
Ink-2Cartesia
8
Qwen3-ASR 0.6BAlibaba
8.7
Voxtral Small Transcribe 3.0Mistral
9.12
FluxDeepgram
10
Granite Speech 5.0 (2B)IBM
11.07
Scribe v2 RealtimeElevenLabs
12
U3.5 Pro RealtimeAssemblyAI
13
02468101214
We ran it, official AppTek scorer
From the AppTek paper (accent average)
Live (streaming), from Cartesia’s Ink-2 post, rounded

Accuracy for its size

On the Open ASR Leaderboard multilingual track, Rumble-2-Flash averages a 6.96% error rate across Chinese, German, French, Italian, Spanish, Portuguese, Hindi, Dutch and English. That's better than Meta's omniASR-LLM 7B, a model 15x larger, and ahead of Whisper large-v3. On the English track it averages 6.6% WER.

11 open-weight models, 9 languages (1 above 12% not shown). Average error rate against model size. Higher up is lower error.

4%6%8%10%12%0.5B1B2B5B10B20B40BQwen3-ASR 1.7BAlibaba · 2.04BQwen3-ASR 0.6BAlibaba · 0.78BNemotron 3.5 TranscribeNVIDIA · 33.7BomniASR-LLM 7BMeta · 7.8BWhisper large-v3OpenAI · 1.55BRumble-2-Flash Open0.5B · 6.96%
Model size, billions of parameters (log scale)– – Best score for each size    ● Other open models

Average over Chinese, German, French, Italian, Spanish, Portuguese, Hindi, Dutch and English.
Hugging Face Open ASR Leaderboard, 13 November 2026. Chinese is scored by character error rate.

The best score for its size

Speed

Rumble-2-Flash is also fast. Its throughput depends on how full the batches are: on the larger datasets in HF's Open ASR Leaderboard, where batches stay full, it reaches over 7,000 RTFx. That's an hour of audio in about half a second.

Languages around the globe

Large models are mostly trained on English and a handful of major languages, and their error rates climb quickly outside that group.

We compared Rumble-2-Flash with OpenAI’s Whisper Large V3, a model three times its size, on ten of these languages. Rumble-2-Flash averages 17.6% WER, compared with 39.8% for Whisper, which is less than half the errors. The gap is widest where other models struggle most: 13% vs. 72% in Maltese, 26% vs. 74% in Bangla, and 13% vs. 42% in Basque.

Rumble-2-Flash Open, 0.5BWhisper Large V3, 1.55B · Word error rate, lower is better
Maltese
13%72%
Bangla
26%74%
Albanian
22%52%
Basque
13%42%
Afrikaans
20%33%
Lithuanian
18%28%
Icelandic
26%36%
Latvian
12%20%
Macedonian
12%18%
Croatian
10%12%
0%10%20%30%40%50%60%70%80%

What you need to know about benchmarking

Speech tests compare what a model writes with what a person typed out, and people make mistakes too. They mishear, or write “twenty” where the model wrote “20”, and the test counts that as wrong.

That’s how anyone’s scores should be read, ours included. If you’re evaluating speech-to-text for your own product, test it on your own audio.

The model is yours.
Are you ready 2 Rumble?

1

Run it

Laptop, on-prem server, device or your own cloud. Open weights so audio never has to leave your hardware.

2

Own it

Licensed under Apache 2.0. Use it commercially, build it into your product.

3

Shape it

Fine-tune for your needs, accents, terminology or language. Small enough to train without a cluster.

For the rooms where every word matters

Even in a busy meeting or a noisy room, Rumble-2-Flash open weights gets around 9 in 10 words right. That’s plenty for a voice note. In a doctor’s appointment, a board meeting or a voice agent serving millions of callers, one wrong word in that last 1 in 10 can change a diagnosis or a decision. That’s what Rumble-2 is for: the most accurate model we’ve trained so far.

Rumble-2
Speaker diarization
End-of-turn detection
Automatic language detection
Real time and prerecorded audio
Tailored to your vocabulary (up to 50K terms)
Our own GPUs in the Netherlands
Zero Data Retention
No training on customer data

voice ai
that speaks your language

Download Rumble-⁠2-⁠Flash
Contact sales