Far-Field ASR Leaderboard
Models transcribe speech under simulated room acoustics. The board compares word error rates.
Treble Technologies and Hugging Face publish the board together.
- Made by
- Treble Technologies and Hugging Face
- Released
- June 2026
- Kind
- Speech recognition and translation
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Short answer
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Items from
- Generated
- Refresh
- One fixed set
- Run by
- A host
- In views
- Audio and speech
- Last model added
- October 2026
- Scores refreshed
- November 20th, 2026
rohansheth/tiro-qwen3-asr-1.7b-v2 tops the leaderboard with 12% Qwen3-ASR-1.7B leads with 13.4%
Best score per entry from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. The best model we track is Qwen3-ASR-1.7B, at 13.4%. This view shows only models we track, each at its best.
- 1 rohansheth/tiro-qwen3-asr-1.7b-v2 12%
- 2 Musci-research/Musci-ASR-qwen-1019-hf 12.8%
- 3 zhifeixie/Mega-ASR 13.4%
- 4
Qwen3-ASR-1.7B 13.4%
- 5 netease-youdao/Confucius4-R2T2 14.2%
- 6 TheStageAI/thewhisper-large-v3-turbo 14.5%
- 7 ibm-granite/granite-speech-5.0-2b 14.8%
- 8
Cohere Transcribe 15%
- 9 ibm-granite/granite-speech-4.1-2b-nar 15.2%
- 10 rohansheth/tiro-qwen3-asr-1.7b-ffasr-v1 15.3%
- 11 ibm-granite/granite-speech-5.0-470m-turboctc-nc 15.8%
- 12 ibm-granite/granite-speech-4.1-2b 16.4%
- 13 salesforce-sovereign-intelligence/SOV-Transcribe-1 16.4%
- 14 FluidVoice-Labs/fluid-speech-pro 16.5%
- 15
Qwen3-ASR-0.6B 17%
- 16 nvidia/parakeet-tdt-1.1b 17.4%
- 17 nvidia/parakeet-tdt-0.6b-v2 17.5%
- 18 ibm-granite/granite-speech-5.0-470m-turboctc 17.6%
- 19 Whisper large-v3 18%
- 20 LelandHenry/Labrador-v0.1 18.2%
- 21
NVIDIA Nemotron 3.5 Transcribe 30B-A4B 18.2%
- 22 nvidia/parakeet-tdt-0.6b-v3 18.3%
- 23 AutoArk-AI/ARK-ASR-3B 19.1%
- 24 ibm-granite/granite-4.0-1b-speech 19.2%
- 25 nvidia/canary-qwen-2.5b 19.5%
- 26 bosonai/higgs-audio-v3-stt 19.7%
- 27 HojoAI/Hojo-ASR-V1 20.1%
- 28 BuzzASR/english 20.4%
- 29 FunAudioLLM/Fun-ASR-Nano-2512 20.4%
- 30 LiquidAI/LFM3-Audio-2B 20.5%
- 31 bosonai/higgs-audio-v3-8b-stt-v2 20.5%
- 32 FermionResearch/Phonon-2 21.1%
- 33 mistralai/Voxtral-Mini-3B-2507 21.6%
- 34 nvidia/canary-1b-v2 21.6%
- 35 frankyoujian/EdgeASR 21.7%
- 36 kienpt0901/ffasr-bd6-rank32-beam-adapter 23.1%
- 37 banhchungtuongot/ffasr-base-beam4-decode 23.3%
- 38 nvidia/canary-1b-flash 24%
- 39 kienpt0901/ffasr-bd6-rank32-adapter-weights 24%
- 40 banhchungtuongot/ffasr-pilot-x-adapter-weights 24.1%
- 41 nvidia/parakeet-ctc-1.1b 24.6%
- 42 FunAudioLLM/SenseVoiceSmall 25.4%
- 43 efficient-speech/lite-whisper-large-v3-turbo-acc 26%
- 44 espnet/owsm_ctc_v4_1B 28%
- 45 Edge0/Audio8-ASR-0.1B 29%
- 46 futo-org/asr4all-l 29.8%
- 47 futo-org/asr4all-l-plus 29.8%
- 48 speechbrain/asr-conformer-loquacious 30.3%
- 49
Voxtral Mini Transcribe Realtime (2602) 30.6%
- 50 GilgameshWind/X-ASR-zh-en 31.5%
- 51 abr-ai/niagara-84m-batch.en 35.2%
- 52 nvidia/stt_en_conformer_ctc_large 35.9%
- 53 efficient-speech/lite-whisper-large-v3-turbo 36.6%
- 54 futo-org/asr4all-s 42.3%
- 55 nvidia/stt_en_fastconformer_hybrid_large_streaming_multi 47.2%
- 56 UsefulSensors/moonshine-streaming-small 50.3%
- 57 openai/whisper-base 57.5%
- 58 microsoft/VibeVoice-ASR 58%
- 59 speechbrain/asr-wav2vec2-librispeech 58.2%
- 60
Gemma 4 12B 59.6%
- 61 openai/whisper-tiny 72.3%
Source: Official Far-Field ASR Leaderboard leaderboard
- 1
Qwen3-ASR-1.7B 13.4%
- 2
Cohere Transcribe 15%
- 3
Qwen3-ASR-0.6B 17%
- 4 Whisper large-v3 18%
- 5
NVIDIA Nemotron 3.5 Transcribe 30B-A4B 18.2%
- 6
Voxtral Mini Transcribe Realtime (2602) 30.6%
- 7
Gemma 4 12B 59.6%
Source: Official Far-Field ASR Leaderboard leaderboard
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | rohansheth/tiro-qwen3-asr-1.7b-v2 2B | 12% | Official leaderboard | ||
| 2 | Musci-research/Musci-ASR-qwen-1019-hf 2B | 12.8% | Official leaderboard | ||
| 3 | zhifeixie/Mega-ASR | 13.4% | Official leaderboard | ||
| 4 | | 13.4% | Jan 2026 | Official leaderboard | |
| 5 | netease-youdao/Confucius4-R2T2 2B | 14.2% | Official leaderboard | ||
| 6 | TheStageAI/thewhisper-large-v3-turbo 0.8B | 14.5% | Official leaderboard | ||
| 7 | ibm-granite/granite-speech-5.0-2b 2.1B | 14.8% | Official leaderboard | ||
| 8 | | 15% | Mar 2026 | Official leaderboard | |
| 9 | ibm-granite/granite-speech-4.1-2b-nar 2.3B | 15.2% | Official leaderboard | ||
| 10 | rohansheth/tiro-qwen3-asr-1.7b-ffasr-v1 2B | 15.3% | Official leaderboard | ||
| 11 | ibm-granite/granite-speech-5.0-470m-turboctc-nc 0.5B | 15.8% | Official leaderboard | ||
| 12 | ibm-granite/granite-speech-4.1-2b 2.3B | 16.4% | Official leaderboard | ||
| 13 | salesforce-sovereign-intelligence/SOV-Transcribe-1 25.9B | 16.4% | Official leaderboard | ||
| 14 | FluidVoice-Labs/fluid-speech-pro 0.6B | 16.5% | Official leaderboard | ||
| 15 | | 17% | Jan 2026 | Official leaderboard | |
| 16 | nvidia/parakeet-tdt-1.1b 1.1B | 17.4% | Official leaderboard | ||
| 17 | nvidia/parakeet-tdt-0.6b-v2 0.6B | 17.5% | Official leaderboard | ||
| 18 | ibm-granite/granite-speech-5.0-470m-turboctc 0.5B | 17.6% | Official leaderboard | ||
| 19 | Whisper large-v3 | 18% | Nov 2023 | Official leaderboard | |
| 20 | LelandHenry/Labrador-v0.1 2B | 18.2% | Official leaderboard | ||
| 21 | | 18.2% | Oct 2026 | Official leaderboard | |
| 22 | nvidia/parakeet-tdt-0.6b-v3 0.6B | 18.3% | Official leaderboard | ||
| 23 | AutoArk-AI/ARK-ASR-3B 3.1B | 19.1% | Official leaderboard | ||
| 24 | ibm-granite/granite-4.0-1b-speech 2.3B | 19.2% | Official leaderboard | ||
| 25 | nvidia/canary-qwen-2.5b 2.6B | 19.5% | Official leaderboard | ||
| 26 | bosonai/higgs-audio-v3-stt 2.7B | 19.7% | Official leaderboard | ||
| 27 | HojoAI/Hojo-ASR-V1 5.2B | 20.1% | Official leaderboard | ||
| 28 | BuzzASR/english 1.6B | 20.4% | Official leaderboard | ||
| 29 | FunAudioLLM/Fun-ASR-Nano-2512 0.8B | 20.4% | Official leaderboard | ||
| 30 | LiquidAI/LFM3-Audio-2B 2B | 20.5% | Official leaderboard | ||
| 31 | bosonai/higgs-audio-v3-8b-stt-v2 8.9B | 20.5% | Official leaderboard | ||
| 32 | FermionResearch/Phonon-2 0.6B | 21.1% | Official leaderboard | ||
| 33 | mistralai/Voxtral-Mini-3B-2507 4.3B | 21.6% | Official leaderboard | ||
| 34 | nvidia/canary-1b-v2 1B | 21.6% | Official leaderboard | ||
| 35 | frankyoujian/EdgeASR 0.1B | 21.7% | Official leaderboard | ||
| 36 | kienpt0901/ffasr-bd6-rank32-beam-adapter 0.6B | 23.1% | Official leaderboard | ||
| 37 | banhchungtuongot/ffasr-base-beam4-decode 0.6B | 23.3% | Official leaderboard | ||
| 38 | nvidia/canary-1b-flash | 24% | Official leaderboard | ||
| 39 | kienpt0901/ffasr-bd6-rank32-adapter-weights 0.6B | 24% | Official leaderboard | ||
| 40 | banhchungtuongot/ffasr-pilot-x-adapter-weights 0.6B | 24.1% | Official leaderboard | ||
| 41 | nvidia/parakeet-ctc-1.1b 1.1B | 24.6% | Official leaderboard | ||
| 42 | FunAudioLLM/SenseVoiceSmall 0.2B | 25.4% | Official leaderboard | ||
| 43 | efficient-speech/lite-whisper-large-v3-turbo-acc 0.6B | 26% | Official leaderboard | ||
| 44 | espnet/owsm_ctc_v4_1B | 28% | Official leaderboard | ||
| 45 | Edge0/Audio8-ASR-0.1B 0.3B | 29% | Official leaderboard | ||
| 46 | futo-org/asr4all-l 0.1B | 29.8% | Official leaderboard | ||
| 47 | futo-org/asr4all-l-plus 0.1B | 29.8% | Official leaderboard | ||
| 48 | speechbrain/asr-conformer-loquacious 0.5B | 30.3% | Official leaderboard | ||
| 49 | | 30.6% | Feb 2026 | Official leaderboard | |
| 50 | GilgameshWind/X-ASR-zh-en | 31.5% | Official leaderboard | ||
| 51 | abr-ai/niagara-84m-batch.en 0.1B | 35.2% | Official leaderboard | ||
| 52 | nvidia/stt_en_conformer_ctc_large 0.1B | 35.9% | Official leaderboard | ||
| 53 | efficient-speech/lite-whisper-large-v3-turbo 0.6B | 36.6% | Official leaderboard | ||
| 54 | futo-org/asr4all-s 0B | 42.3% | Official leaderboard | ||
| 55 | nvidia/stt_en_fastconformer_hybrid_large_streaming_multi 0.1B | 47.2% | Official leaderboard | ||
| 56 | UsefulSensors/moonshine-streaming-small 0.1B | 50.3% | Official leaderboard | ||
| 57 | openai/whisper-base 0.1B | 57.5% | Official leaderboard | ||
| 58 | microsoft/VibeVoice-ASR 8.7B | 58% | Official leaderboard | ||
| 59 | speechbrain/asr-wav2vec2-librispeech 0.3B | 58.2% | Official leaderboard | ||
| 60 | | 59.6% | Jun 2026 | Official leaderboard | |
| 61 | openai/whisper-tiny 0B | 72.3% | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 13.4% | Jan 2026 | Official leaderboard | |
| 2 | | 15% | Mar 2026 | Official leaderboard | |
| 3 | | 17% | Jan 2026 | Official leaderboard | |
| 4 | Whisper large-v3 | 18% | Nov 2023 | Official leaderboard | |
| 5 | | 18.2% | Oct 2026 | Official leaderboard | |
| 6 | | 30.6% | Feb 2026 | Official leaderboard | |
| 7 | | 59.6% | Jun 2026 | Official leaderboard |
Rankings by part
We read 9 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Near Field, Lab Meas. and Lab Sim.. Qwen3-ASR-1.7B leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.
Near Field
7 models we track have a score here.
- 1
Qwen3-ASR-1.7B 3.76%
- 2
NVIDIA Nemotron 3.5 Transcribe 30B-A4B ↑3 3.81%
- 3 Whisper large-v3 ↑1 3.87%
- 4
Cohere Transcribe ↓2 4.28%
- 5
Qwen3-ASR-0.6B ↓2 4.37%
- 6
Voxtral Mini Transcribe Realtime (2602) 4.75%
- 7
Gemma 4 12B 6.24%
Lab Meas.
7 models we track have a score here.
- 1
Qwen3-ASR-1.7B 19.9%
- 2
Cohere Transcribe 21.4%
- 3
Qwen3-ASR-0.6B 25.2%
- 4 Whisper large-v3 25.3%
- 5
NVIDIA Nemotron 3.5 Transcribe 30B-A4B 26.4%
- 6
Voxtral Mini Transcribe Realtime (2602) 47.3%
- 7
Gemma 4 12B 79.9%
Lab Sim.
7 models we track have a score here.
- 1
Qwen3-ASR-1.7B 18.6%
- 2
Cohere Transcribe 20.2%
- 3
Qwen3-ASR-0.6B 24.1%
- 4 Whisper large-v3 24.4%
- 5
NVIDIA Nemotron 3.5 Transcribe 30B-A4B 24.9%
- 6
Voxtral Mini Transcribe Realtime (2602) 45.3%
- 7
Gemma 4 12B 87%
High SNR
7 models we track have a score here.
- 1
Qwen3-ASR-1.7B 6.74%
- 2
Cohere Transcribe 7.79%
- 3
NVIDIA Nemotron 3.5 Transcribe 30B-A4B ↑2 8.86%
- 4 Whisper large-v3 8.98%
- 5
Qwen3-ASR-0.6B ↓2 9.59%
- 6
Voxtral Mini Transcribe Realtime (2602) 18.8%
- 7
Gemma 4 12B 26.5%
Mid SNR
7 models we track have a score here.
- 1
Qwen3-ASR-1.7B 13.9%
- 2
Cohere Transcribe 15.3%
- 3
Qwen3-ASR-0.6B 18.2%
- 4
NVIDIA Nemotron 3.5 Transcribe 30B-A4B ↑1 18.4%
- 5 Whisper large-v3 ↓1 18.8%
- 6
Voxtral Mini Transcribe Realtime (2602) 36.7%
- 7
Gemma 4 12B 68.3%
Low SNR
7 models we track have a score here.
- 1
Qwen3-ASR-1.7B 29.3%
- 2
Cohere Transcribe 32.8%
- 3
Qwen3-ASR-0.6B 35.7%
- 4 Whisper large-v3 40.5%
- 5
NVIDIA Nemotron 3.5 Transcribe 30B-A4B 41.8%
- 6
Voxtral Mini Transcribe Realtime (2602) 62.4%
- 7
Gemma 4 12B 137.3%
Mov. High
7 models we track have a score here.
- 1
Qwen3-ASR-1.7B 8.17%
- 2
Cohere Transcribe 8.81%
- 3
NVIDIA Nemotron 3.5 Transcribe 30B-A4B ↑2 9.78%
- 4 Whisper large-v3 10%
- 5
Qwen3-ASR-0.6B ↓2 11.3%
- 6
Voxtral Mini Transcribe Realtime (2602) 20.8%
- 7
Gemma 4 12B 29.3%
Mov. Mid
7 models we track have a score here.
- 1
Qwen3-ASR-1.7B 15.9%
- 2
Cohere Transcribe 17%
- 3
NVIDIA Nemotron 3.5 Transcribe 30B-A4B ↑2 20.7%
- 4
Qwen3-ASR-0.6B ↓1 21.2%
- 5 Whisper large-v3 ↓1 21.8%
- 6
Voxtral Mini Transcribe Realtime (2602) 41.9%
- 7
Gemma 4 12B 70.4%
Mov. Low
7 models we track have a score here.
- 1
Qwen3-ASR-1.7B 31.9%
- 2
Cohere Transcribe 33.9%
- 3
Qwen3-ASR-0.6B 38.4%
- 4
NVIDIA Nemotron 3.5 Transcribe 30B-A4B ↑1 42.2%
- 5 Whisper large-v3 ↓1 42.4%
- 6
Voxtral Mini Transcribe Realtime (2602) 64.7%
- 7
Gemma 4 12B 135.1%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official Far-Field ASR Leaderboard leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official Far-Field ASR Leaderboard leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Audio and video
- Video-MMMU In use 300 expert-level lecture videos with questions testing perception, comprehension and adaptation of knowledge. Accuracy.
Qwen 3.8 Max 88.7%
- LVBench In use Hours-long video understanding with multiple-choice questions on reasoning, grounding and summarization.
Gemini 4 Argon 91.7%
- MMVU In use 3,000 expert-annotated questions on videos across 27 disciplines. Multiple-choice and open-ended accuracy.
Qwen 3.8 Max 82.4%
- MLVU In use Multi-task long video understanding (minutes to hours) with multiple-choice and generation tasks. Accuracy.
Qwen 3.8 Max 90.8%
- FLEURS In use Models transcribe read speech in 102 languages built from FLoRes-101 sentences. Model cards usually report word error rate, so lower scores are better.
Phi-4-multimodal 4.00%
- VoiceBench In use Spoken instructions with speaker, environment and content variation. Voice assistant responses scored by GPT judge or rules.
Qwen3.5-Omni-Plus 93.1%
Data audit
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (61 entries).
- Pass Models scored 7 models we track, plus 54 other entries on its leaderboard.
- Pass Newest score The newest is from October 2026, for NVIDIA Nemotron 3.5 Transcribe 30B-A4B.
- Note Frontier models None of its models are on the capabilities index.
- Note Headroom Lower is better, so there’s no ceiling to reach.
Leaderboard import 3 pass
- Pass Importer We read the board on huggingface.co: 7 rows for models we track and 54 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.