Transcription

Speech-recognition models ranked by word error rate, transcription and dictation apps, and open engines that transcribe entirely on your own machine.

Models are ranked by Artificial Analysis Speech to Text: AA-WER v2 (non-streaming), Hugging Face Open ASR Leaderboard (English, short-form average WER), Artificial Analysis Speech to Text: AA-WER Streaming, then Coval Benchmarks: Speech-to-Text WER (24h window) for models without a result on the first. Tools inherit the tier of the best model they offer. How we rank.

Best overallSFun-Realtime-ASR (preview)1.7% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)
Best valueAModulate STT (batch)Tier A for $0.00042/min of audio
Best free startSElevenLabsFree tier · from $6/mo · via Scribe v2
Best local (≤24 GB GPU)AQwen3-ASR-1.7B~6 GB VRAM · 5.4% on Hugging Face Open ASR Leaderboard (English, short-form average WER)

Tier: how close to the best result on the task's main public benchmark
Outlined: a tool inherits the tier of the best model it offers
No public result yet
Cloud · Local: runs on your own hardware

Models

The engines themselves. Use them through an API or inside the apps on the right.

Cloud

Fun-Realtime-ASR (preview)Alibaba1.7% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)S

MAI-Transcribe-2 (preview)Microsoft AI · $0.0017/min of audio2% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)$S

Scribe v2ElevenLabs · $0.0037/min of audio2.2% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)S

Grok Voice Transcribe 2.0xAI · $0.0017/min of audio2.3% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)S

StepAudio 3 ASRStepFun · $0.0067/min of audio2.5% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)S

Gemini 3.5 TranscribeGoogle · $0.0050/min of audio2.6% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

Ink-2Cartesia · $0.0040/min of audio3.4% on Artificial Analysis Speech to Text: AA-WER Streaming A

Scribe V2 ProZoom · $0.0033/min of audio2.8% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

Saaras v4Sarvam AI · $0.0052/min of audio3.5% on Artificial Analysis Speech to Text: AA-WER Streaming A

Universal-3.6 Pro StreamingAssemblyAI · $0.0075/min of audio3.6% on Artificial Analysis Speech to Text: AA-WER Streaming A

Gemini 3.1 Pro (Preview)Google · $0.02/min of audio2.8% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

Universal-3.5 ProAssemblyAI · $0.0035/min of audio3% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

Solaria-3Gladia · $0.01/min of audio3.2% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

GPT-TranscribeOpenAI · $0.0045/min of audio3.3% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

PulseSmallest.ai · $0.0050/min of audio3.3% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

Resonant-1Reson8 · $0.0036/min of audio3.4% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

Muse Voice TranscribeMeta · $0.0030/min of audio6.3% on Hugging Face Open ASR Leaderboard (English, short-form average WER)A

Qwen3.5-Omni-PlusAlibaba (Qwen)3.5% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

Voxtral Mini Transcribe 2 (v26.02)Mistral AI · $0.0030/min of audio3.6% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

Soniox v5 (stt-rt-v5 / async)Soniox · $0.0017/min of audio3.8% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

Inworld STT 1Inworld AI · $0.0025/min of audio3.9% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

Amazon TranscribeAmazon · $0.0060/min of audio4.1% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

Modulate STT (batch)Modulate · $0.00042/min of audio4.2% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)$B

Chirp 3Google · $0.02/min of audio4.3% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)B

Nova-3Deepgram · $0.0043/min of audio4.8% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)B

Melia 1Speechmatics · $0.0040/min of audio4.9% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)C

Flux (English / Multilingual)Deepgram · $0.0065/min of audio7.4% on Artificial Analysis Speech to Text: AA-WER Streaming C

Gradium Speech-to-TextGradium · $0.01/min of audio6.8% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)

Local

Qwen3-ASR-1.7BAlibaba (Qwen) · ~6 GB VRAM5.4% on Hugging Face Open ASR Leaderboard (English, short-form average WER)$A

Voxtral Small (24B-2507)Mistral AI · ~55 GB VRAM2.8% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)A

Higgs Audio v3 STTBoson AI · ~6 GB VRAM5.9% on Hugging Face Open ASR Leaderboard (English, short-form average WER)A

Granite Speech 4.1 2BIBM · ~6 GB VRAM6% on Hugging Face Open ASR Leaderboard (English, short-form average WER)A

MOSS-Transcribe-DiarizeOpenMOSS (Fudan University / MOSI) · ~2 GB VRAM6.2% on Hugging Face Open ASR Leaderboard (English, short-form average WER)$A

distil-large-v3.5Hugging Face · ~2 GB VRAM6.8% on Hugging Face Open ASR Leaderboard (English, short-form average WER)A

Whisper large-v3OpenAI · ~10 GB VRAM4.1% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)B

Canary-Qwen-2.5BNVIDIA · ~6 GB VRAM4.3% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)B

Fun-ASR-Nano-2512Alibaba (FunAudioLLM) · ~2 GB VRAM7.2% on Hugging Face Open ASR Leaderboard (English, short-form average WER)B

Parakeet TDT 0.6B v3NVIDIA · ~2 GB VRAM4.5% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)B

cohere-transcribe-03-2026Cohere · ~6 GB VRAM4.6% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)B

omniASR-LLM-7B-v2Meta (FAIR) · ~16 GB VRAM8.1% on Hugging Face Open ASR Leaderboard (English, short-form average WER)B

VibeVoice-ASRMicrosoft · ~19 GB VRAM8.2% on Hugging Face Open ASR Leaderboard (English, short-form average WER)B

Voxtral Mini 4B Realtime (2602)Mistral AI · ~10 GB VRAM8.4% on Hugging Face Open ASR Leaderboard (English, short-form average WER)C

Nemotron 3.5 ASR Streaming 0.6BNVIDIA · ~2 GB VRAM9.8% on Hugging Face Open ASR Leaderboard (English, short-form average WER)–

MiMo-V2.5-ASRXiaomi · ~17 GB VRAMNo public result yet –

Moonshine StreamingMoonshine AI (formerly Useful Sensors) · Open weightsNo public result yet –

stt-2.6b-enKyutai · ~6 GB VRAMNo public result yet

Apps

Ready-made apps with a UI: sign in or install, then start.

Cloud

ElevenLabsElevenLabs · Free tier · $6/movia Scribe v2$A

TurboScribeTurboScribe · Free tier · $20/movia Whisper large-v3 –

Aqua VoiceAqua Voice · Free tier · $8/moNo public result yet –

DescriptDescript · Free tier · $24/moNo public result yet –

NottaNotta · Free tier · $8.17/moNo public result yet –

Otter.aiOtter.ai · Free tier · $17/moNo public result yet –

RevRev.com · Free tier · $30/moNo public result yet –

SonixSonix · Free tier · $25/moNo public result yet –

TrintTrintNo public result yet –

WillowWillow · Free tier · $12/moNo public result yet –

Wispr FlowWispr · Free tier · $15/moNo public result yet

Local

VoiceboxJamie Pine (open source) · Freevia Whisper large-v3$A

HandyCJ Pais (open source) · Freevia Whisper large-v3$A

BuzzChidi Williams (open source) · Freevia Whisper large-v3$A Vibethewh1teagle (open source) · Freevia Whisper large-v3$A

VoiceInkPrakash Joshi Pax (open source)via Whisper large-v3 A Whishperpluja (open source) · Freevia Whisper large-v3$A

AikoSindre Sorhus · Free tiervia Whisper large-v3 A

MacWhisperJordi Bruin (Good Snooze) · Free tiervia Whisper large-v3 A

SpokenlySpokenly · Freevia Whisper large-v3$A

SuperwhisperSuperwhisper · Free tier · $8.49/movia Whisper large-v3

Engines & APIs

Run open speech-recognition models locally, or call transcription APIs from your own apps.

Cloud

Nothing tracked yet

Local

whisper.cppggml-org (open source) · Freevia Whisper large-v3$A

faster-whisperSYSTRAN (open source) · Freevia Whisper large-v3$A

WhisperXMax Bain (open source) · Freevia Whisper large-v3$A sherpa-onnxk2-fsa (Next-gen Kaldi) · Freevia Whisper large-v3$A

WhisperKitArgmax · Freevia Whisper large-v3$A

WhisperLiveCollabora · Freevia Whisper large-v3$A

Speachesspeaches-ai (open source) · Freevia Whisper large-v3$B

NVIDIA NeMo SpeechNVIDIA · Freevia Canary-Qwen-2.5B –

VoskAlpha Cephei · FreeNo public result yet –

Moonshine VoiceMoonshine AI · FreeNo public result yet

Quality vs. price

Artificial Analysis Speech to Text: AA-WER v2 (non-streaming) score against API price (USD per minute of audio, log scale). Up and to the left is better; the line joins the models that nothing else beats on both.

Closed models
Open-weights models
Best-value frontier

Show as table

Model	Weights	Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	Price	Frontier
MAI-Transcribe-2 (preview)	Closed	2%	$0.0017/min of audio	Yes
Scribe v2	Closed	2.2%	$0.0037/min of audio
Grok Voice Transcribe 2.0	Closed	2.3%	$0.0017/min of audio
StepAudio 3 ASR	Closed	2.5%	$0.0067/min of audio
Gemini 3.5 Transcribe	Closed	2.6%	$0.0050/min of audio
Scribe V2 Pro	Closed	2.8%	$0.0033/min of audio
Voxtral Small (24B-2507)	Open	2.8%	$0.0040/min of audio
Gemini 3.1 Pro (Preview)	Closed	2.8%	$0.02/min of audio
Universal-3.5 Pro	Closed	3%	$0.0035/min of audio
Solaria-3	Closed	3.2%	$0.01/min of audio
GPT-Transcribe	Closed	3.3%	$0.0045/min of audio
Pulse	Closed	3.3%	$0.0050/min of audio
Resonant-1	Closed	3.4%	$0.0036/min of audio
Voxtral Mini Transcribe 2 (v26.02)	Closed	3.6%	$0.0030/min of audio
Soniox v5 (stt-rt-v5 / async)	Closed	3.8%	$0.0017/min of audio
Inworld STT 1	Closed	3.9%	$0.0025/min of audio
Whisper large-v3	Open	4.1%	$0.0011/min of audio	Yes
Amazon Transcribe	Closed	4.1%	$0.0060/min of audio
Modulate STT (batch)	Closed	4.2%	$0.00042/min of audio	Yes
Canary-Qwen-2.5B	Open	4.3%	$0.00074/min of audio
Chirp 3	Closed	4.3%	$0.02/min of audio
Parakeet TDT 0.6B v3	Open	4.5%	$0.0015/min of audio
Nova-3	Closed	4.8%	$0.0043/min of audio
Melia 1	Closed	4.9%	$0.0040/min of audio
Gradium Speech-to-Text	Closed	6.8%	$0.01/min of audio

Everything, in order

All transcription options in quality order
Tier	Name	Approach	Runs	Rated by	Price	Skill
S	Fun-Realtime-ASR (preview) Alibaba	Models	Cloud	1.7% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	—	—
S	MAI-Transcribe-2 (preview) Microsoft AI	Models	Cloud	2% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0017/min of audio	—
S	Scribe v2 ElevenLabs	Models	Cloud	2.2% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0037/min of audio	—
S	Grok Voice Transcribe 2.0 xAI	Models	Cloud	2.3% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0017/min of audio	—
S	StepAudio 3 ASR StepFun	Models	Cloud	2.5% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0067/min of audio	—
S	Gemini 3.5 Transcribe Google	Models	Cloud	2.6% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0050/min of audio	—
S	ElevenLabs ElevenLabs	Apps	Cloud	via Scribe v2	Free tier · $6/mo	beginner
A	Ink-2 Cartesia	Models	Cloud	3.4% on Artificial Analysis Speech to Text: AA-WER Streaming	$0.0040/min of audio	—
A	Qwen3-ASR-1.7B Alibaba (Qwen)	Models	Local	5.4% on Hugging Face Open ASR Leaderboard (English, short-form average WER)	~6 GB VRAM	—
A	Scribe V2 Pro Zoom	Models	Cloud	2.8% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0033/min of audio	—
A	Saaras v4 Sarvam AI	Models	Cloud	3.5% on Artificial Analysis Speech to Text: AA-WER Streaming	$0.0052/min of audio	—
A	Voxtral Small (24B-2507) Mistral AI	Models	Local	2.8% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	~55 GB VRAM	—
A	Universal-3.6 Pro Streaming AssemblyAI	Models	Cloud	3.6% on Artificial Analysis Speech to Text: AA-WER Streaming	$0.0075/min of audio	—
A	Gemini 3.1 Pro (Preview) Google	Models	Cloud	2.8% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.02/min of audio	—
A	Universal-3.5 Pro AssemblyAI	Models	Cloud	3% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0035/min of audio	—
A	Higgs Audio v3 STT Boson AI	Models	Local	5.9% on Hugging Face Open ASR Leaderboard (English, short-form average WER)	~6 GB VRAM	—
A	Granite Speech 4.1 2B IBM	Models	Local	6% on Hugging Face Open ASR Leaderboard (English, short-form average WER)	~6 GB VRAM	—
A	Solaria-3 Gladia	Models	Cloud	3.2% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.01/min of audio	—
A	GPT-Transcribe OpenAI	Models	Cloud	3.3% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0045/min of audio	—
A	Pulse Smallest.ai	Models	Cloud	3.3% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0050/min of audio	—
A	MOSS-Transcribe-Diarize OpenMOSS (Fudan University / MOSI)	Models	Local	6.2% on Hugging Face Open ASR Leaderboard (English, short-form average WER)	~2 GB VRAM	—
A	Resonant-1 Reson8	Models	Cloud	3.4% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0036/min of audio	—
A	Muse Voice Transcribe Meta	Models	Cloud	6.3% on Hugging Face Open ASR Leaderboard (English, short-form average WER)	$0.0030/min of audio	—
A	Qwen3.5-Omni-Plus Alibaba (Qwen)	Models	Cloud	3.5% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	—	—
A	Voxtral Mini Transcribe 2 (v26.02) Mistral AI	Models	Cloud	3.6% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0030/min of audio	—
A	Soniox v5 (stt-rt-v5 / async) Soniox	Models	Cloud	3.8% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0017/min of audio	—
A	Inworld STT 1 Inworld AI	Models	Cloud	3.9% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0025/min of audio	—
A	distil-large-v3.5 Hugging Face	Models	Local	6.8% on Hugging Face Open ASR Leaderboard (English, short-form average WER)	~2 GB VRAM	—
A	Whisper large-v3 OpenAI	Models	Local	4.1% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	~10 GB VRAM	—
A	Amazon Transcribe Amazon	Models	Cloud	4.1% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0060/min of audio	—
A	Modulate STT (batch) Modulate	Models	Cloud	4.2% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.00042/min of audio	—
A	Voicebox Jamie Pine (open source)	Apps	Local	via Whisper large-v3	Free	beginner
A	whisper.cpp ggml-org (open source)	Engines & APIs	Local	via Whisper large-v3	Free	advanced
A	Handy CJ Pais (open source)	Apps	Local	via Whisper large-v3	Free	beginner
A	faster-whisper SYSTRAN (open source)	Engines & APIs	Local	via Whisper large-v3	Free	advanced
A	WhisperX Max Bain (open source)	Engines & APIs	Local	via Whisper large-v3	Free	advanced
A	Buzz Chidi Williams (open source)	Apps	Local	via Whisper large-v3	Free	beginner
A	sherpa-onnx k2-fsa (Next-gen Kaldi)	Engines & APIs	Local	via Whisper large-v3	Free	advanced
A	Vibe thewh1teagle (open source)	Apps	Local	via Whisper large-v3	Free	beginner
A	VoiceInk Prakash Joshi Pax (open source)	Apps	Local	via Whisper large-v3	—	beginner
A	WhisperKit Argmax	Engines & APIs	Local	via Whisper large-v3	Free	advanced
A	WhisperLive Collabora	Engines & APIs	Local	via Whisper large-v3	Free	advanced
A	Speaches speaches-ai (open source)	Engines & APIs	Local	via Whisper large-v3	Free	advanced
A	Whishper pluja (open source)	Apps	Local	via Whisper large-v3	Free	advanced
A	Aiko Sindre Sorhus	Apps	Local	via Whisper large-v3	Free tier	beginner
A	MacWhisper Jordi Bruin (Good Snooze)	Apps	Local	via Whisper large-v3	Free tier	beginner
A	Spokenly Spokenly	Apps	Local	via Whisper large-v3	Free	beginner
A	Superwhisper Superwhisper	Apps	Local	via Whisper large-v3	Free tier · $8.49/mo	beginner
A	TurboScribe TurboScribe	Apps	Cloud	via Whisper large-v3	Free tier · $20/mo	beginner
B	Canary-Qwen-2.5B NVIDIA	Models	Local	4.3% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	~6 GB VRAM	—
B	Chirp 3 Google	Models	Cloud	4.3% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.02/min of audio	—
B	Fun-ASR-Nano-2512 Alibaba (FunAudioLLM)	Models	Local	7.2% on Hugging Face Open ASR Leaderboard (English, short-form average WER)	~2 GB VRAM	—
B	Parakeet TDT 0.6B v3 NVIDIA	Models	Local	4.5% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	~2 GB VRAM	—
B	cohere-transcribe-03-2026 Cohere	Models	Local	4.6% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	~6 GB VRAM	—
B	Nova-3 Deepgram	Models	Cloud	4.8% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0043/min of audio	—
B	Melia 1 Speechmatics	Models	Cloud	4.9% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.0040/min of audio	—
B	omniASR-LLM-7B-v2 Meta (FAIR)	Models	Local	8.1% on Hugging Face Open ASR Leaderboard (English, short-form average WER)	~16 GB VRAM	—
B	VibeVoice-ASR Microsoft	Models	Local	8.2% on Hugging Face Open ASR Leaderboard (English, short-form average WER)	~19 GB VRAM	—
B	Voxtral Mini 4B Realtime (2602) Mistral AI	Models	Local	8.4% on Hugging Face Open ASR Leaderboard (English, short-form average WER)	~10 GB VRAM	—
B	NVIDIA NeMo Speech NVIDIA	Engines & APIs	Local	via Canary-Qwen-2.5B	Free	advanced
C	Flux (English / Multilingual) Deepgram	Models	Cloud	7.4% on Artificial Analysis Speech to Text: AA-WER Streaming	$0.0065/min of audio	—
C	Gradium Speech-to-Text Gradium	Models	Cloud	6.8% on Artificial Analysis Speech to Text: AA-WER v2 (non-streaming)	$0.01/min of audio	—
C	Nemotron 3.5 ASR Streaming 0.6B NVIDIA	Models	Local	9.8% on Hugging Face Open ASR Leaderboard (English, short-form average WER)	~2 GB VRAM	—
–	Vosk Alpha Cephei	Engines & APIs	Local	No public result yet	Free	advanced
–	Moonshine Voice Moonshine AI	Engines & APIs	Local	No public result yet	Free	advanced
–	Aqua Voice Aqua Voice	Apps	Cloud	No public result yet	Free tier · $8/mo	beginner
–	Descript Descript	Apps	Cloud	No public result yet	Free tier · $24/mo	beginner
–	MiMo-V2.5-ASR Xiaomi	Models	Local	No public result yet	~17 GB VRAM	—
–	Moonshine Streaming Moonshine AI (formerly Useful Sensors)	Models	Local	No public result yet	Open weights	—
–	Notta Notta	Apps	Cloud	No public result yet	Free tier · $8.17/mo	beginner
–	Otter.ai Otter.ai	Apps	Cloud	No public result yet	Free tier · $17/mo	beginner
–	Rev Rev.com	Apps	Cloud	No public result yet	Free tier · $30/mo	beginner
–	Sonix Sonix	Apps	Cloud	No public result yet	Free tier · $25/mo	beginner
–	stt-2.6b-en Kyutai	Models	Local	No public result yet	~6 GB VRAM	—
–	Trint Trint	Apps	Cloud	No public result yet	—	beginner
–	Willow Willow	Apps	Cloud	No public result yet	Free tier · $12/mo	beginner
–	Wispr Flow Wispr	Apps	Cloud	No public result yet	Free tier · $15/mo	beginner

Benchmarks used

Artificial Analysis Speech to Text: AA-WER v2 (non-streaming) by Artificial Analysis (automated tests, AA-WER v2: word error rate across AA's own test set plus cleaned VoxPopuli and Earnings-22 subsets (non-streaming)) · snapshot 19 Oct 2026
The page states 57 models evaluated; 60 provider/model rows with an aaWerIndex were read from its embedded data. Values converted from fractions to percent. AssemblyAI rows are Universal-3.5 Pro (2.99), Universal-3 Pro (3.12) and Universal (3.82); Universal-3.6 Pro appears only on the streaming board. No last-updated date shown, so as_of = retrieval date.
Hugging Face Open ASR Leaderboard (English, short-form average WER) by Hugging Face (hf-audio) (automated tests, Average word error rate over the default English short-form datasets (AMI, Earnings22, Gigaspeech, LibriSpeech clean/other, SPGISpeech, URGENT24, Voice Arena, VoxPopuli) plus two private sets) · snapshot 17 Oct 2026
56 ranked entries (open models and commercial APIs) read from the Space's initial table data; results version 17-10-2026 (results dataset hf-audio/open-asr-leaderboard-results). Multilingual and long-form views exist separately.
Artificial Analysis Speech to Text: AA-WER Streaming by Artificial Analysis (automated tests, AA-WER Streaming index: word error rate of final streaming transcripts across AA's agent-talk set plus VoxPopuli and Earnings-22 subsets) · snapshot 19 Oct 2026
45 streaming provider/model rows (page states 45 models); values read from the page's embedded data (field aaWerStreamingIndex, converted to percent). Includes Universal-3.6 Pro Realtime (Min Latency 3.57, Max Accuracy 3.85; release date 2026-09-29) and Universal-3.5 Pro variants. Separate from the non-streaming aa-speech-to-text board. Latency is published alongside but not recorded. No last-updated date shown, so as_of = retrieval date.
Coval Benchmarks: Speech-to-Text WER (24h window) by Coval (automated tests, Mean word error rate (Whisper English normaliser) over voice-agent and noisy-condition clips, rolling 24-hour window, re-run every 30 minutes) · snapshot 19 Oct 2026
31 provider/model entries read from the public API behind benchmarks.coval.ai (snapshot generation 2611, as_of 2026-10-19T10:00Z, stale=false). Values move every 30 minutes because the window is rolling; the page itself renders with JavaScript. Methodology (github.com/coval-ai/benchmarks, Apache-2.0) runs AssemblyAI over its WebSocket streaming endpoint, so the AssemblyAI entries (universal-3.6-pro, universal-3.5-pro, universal-streaming) are streaming results. TTS and speech-to-speech boards measure latency only and are not recorded.

Transcription

Quality vs. price

Quality vs. setup

Everything, in order

Benchmarks used