The Best Open Speech Recognition Models in 2026

For years "open speech recognition" meant Whisper. That is no longer true. By September 2026, more than a dozen open-weight models beat Whisper large-v3 on the standard English benchmark, several of them at a fraction of its size, and OpenAI itself has deprecated Whisper in its paid API in favour of a closed successor. This article surveys the open models that matter, what makes each different, and which are practical to run on a laptop or phone.

All accuracy figures are word error rates (WER) from the Hugging Face Open ASR Leaderboard results as of 19 September 2026, unless stated otherwise. Lower is better.

How to read the leaderboard in 2026

The Open ASR Leaderboard averages WER across eight public English test sets: audiobooks, meetings, earnings calls, podcasts, parliamentary speech and more. In 2026 it switched to cleaned versions of several sets, which lowered every model's score. A number on a model card from 2025 is not comparable with the current leaderboard. Parakeet TDT v2's card says 6.05%; on the current leaderboard it is 4.70%. Same model, different scoring.

The leaderboard also reports RTFx, how many seconds of audio a model transcribes per second on a data-centre GPU. It is a relative speed measure, not what you will see on a laptop.

The top open models for English

A selection of the leading open-weight models; a few less widely used entries (Hojo-ASR, Higgs Audio, ARK-ASR, Phi-4-multimodal) also sit in this range.

Model	Maker	Size	WER	Speed (RTFx)	Licence
Qwen3-ASR 1.7B	Alibaba Qwen	1.7B	4.31	820	Apache 2.0
Canary-Qwen 2.5B	NVIDIA	2.5B	4.43		CC-BY-4.0
Granite Speech 4.1 2B	IBM	2B	4.62		Apache 2.0
MOSS-Transcribe-Diarize	OpenMOSS (Fudan)	0.9B	4.64		Apache 2.0
Cohere Transcribe 03-2026	Cohere	2B	4.67		Apache 2.0
Parakeet TDT 0.6B v2	NVIDIA	0.6B	4.70	6,025	CC-BY-4.0
Parakeet TDT 0.6B v3	NVIDIA	0.6B	4.86	6,076	CC-BY-4.0
Voxtral Small	Mistral	24B	4.99		Apache 2.0
Granite Speech 5.0 TurboCTC	IBM	0.47B	5.04	12,946	Apache 2.0
Qwen3-ASR 0.6B	Alibaba Qwen	0.6B	5.05		Apache 2.0
Kyutai STT 2.6B	Kyutai	2.6B	5.57		CC-BY-4.0
VibeVoice-ASR	Microsoft	~9B	5.58		MIT
Whisper large-v3	OpenAI	1.55B	5.78		MIT
Whisper large-v3-turbo	OpenAI	0.8B	6.36		MIT

The top five overall are all commercial APIs (Zoom, Azure, Modulate, ElevenLabs, Reson8), between 3.6% and 4.0%. The best open model is within about 0.7 points of them.

The spread among the top open models is small, less than a percentage point across the first ten. In practice, a difference of 0.3 points on this benchmark is less important than how a model handles your audio, how fast it runs on your hardware, and whether it invents text over silence.

The models worth knowing

Qwen3-ASR (Alibaba, January 2026)

The most accurate open model on English and one of the broadest: 52 languages and dialects. It is built on the audio encoder of Qwen3-Omni with a language-model decoder, and ships with a separate forced aligner for word timestamps in 11 languages. The 0.6B version answers in about 92 ms and scores 5.05, still ahead of Whisper large-v3.

NVIDIA Parakeet TDT v2 and v3 (2025)

No longer the most accurate, but still the best balance of accuracy and speed. At 600 million parameters, both are within about half a point of the leader and around seven times faster than Qwen3-ASR 1.7B. Their Token-and-Duration Transducer decoder predicts each token together with how many audio frames to skip, which is what makes them so fast. Version 3 covers 25 European languages with automatic detection and is also the best NVIDIA model on long recordings. We compare it with Whisper in Whisper vs Parakeet.

A September 2026 derivative, Parakeet Ultra by moondream, is a post-trained version of v3 that cuts multilingual error noticeably (FLEURS 11.67% against 14.81% in FluidAudio's measurements on Apple silicon) with the same architecture.

NVIDIA Canary-Qwen 2.5B and Canary-1B-v2 (2025)

Canary-Qwen pairs a FastConformer speech encoder with a Qwen3 1.7B language model and is the most accurate CC-BY model. Canary-1B-v2 does speech recognition and translation between English and 24 European languages.

NVIDIA Nemotron 3.5 ASR (June 2026)

A 600M streaming model for 40 locales, grouped by how ready each language is. It is cache-aware: each new chunk of audio reuses the model's previous state instead of recomputing overlapping context, which suits live captions. It pairs with the new Nemotron 3 Diarization model for speaker labels.

IBM Granite Speech (2026)

Two very different models. Granite Speech 4.1 2B couples a conformer encoder with a small Granite language model for six languages. Granite Speech 5.0 TurboCTC, released in August 2026, is an English-only 470M model with no language-model decoder at all, just an encoder trained with CTC. It is the fastest model on the leaderboard at nearly 13,000 RTFx and still beats Whisper. IBM pitches it at laptops and phones.

Cohere Transcribe (March 2026)

A 2B model for 14 languages, and the most accurate open model on the leaderboard's long-form track (9.73%). If you mostly transcribe hour-long recordings, that track matters more than short clips.

Mistral Voxtral (2025 and 2026)

Voxtral Small (24B) and Mini (3B) combine transcription with audio understanding: you can ask questions about a recording. Voxtral Transcribe 2, from February 2026, adds an open 4B realtime model with a delay you can set between 80 ms and 2.4 s.

Microsoft VibeVoice-ASR (January 2026) and MOSS-Transcribe-Diarize (2026)

Both are part of 2026's biggest trend: one model that transcribes and labels speakers in a single pass. VibeVoice-ASR takes up to 60 minutes of audio in one 64K-token context and outputs who spoke, when, and what. MOSS-Transcribe-Diarize does the same in 50+ languages at under 1B parameters.

Meta Omnilingual ASR (November 2025)

The widest language coverage by far: more than 1,600 languages, over 500 never covered by any system before, with a character error rate under 10% for 78% of them. It scales a wav2vec 2.0 encoder to 7B parameters and includes a zero-shot variant that learns a new language from a few examples. Not the choice for English accuracy, but unmatched for low-resource languages.

Kyutai STT (2025)

Streams text with a fixed delay by modelling audio and text as parallel token streams. The 1B model supports English and French with half a second of delay.

What happened to Whisper

On English, Whisper large-v3 now ranks below nearly twenty open models. It remains useful for its language breadth (about 99 languages), its translation-to-English mode, and the huge ecosystem around it (whisper.cpp, MLX, Core ML ports). OpenAI released no open successor. On 26 August 2026 it deprecated whisper-1 and its gpt-4o-transcribe models in the API in favour of a new closed gpt-transcribe model, with shutdown scheduled for February 2027. The open Whisper weights remain available.

Closed models set the pace

The fastest movement in 2026 is in commercial APIs: ElevenLabs Scribe v2 leads long-form, Google's Gemini 3.5 Transcribe (August 2026) and Meta's Muse Voice Transcribe (September 2026) add streaming and speaker labels, and OpenAI's gpt-transcribe replaces Whisper in its API. They are accurate, but your audio goes to the provider. For the trade-offs, see on-device vs cloud transcription.

Which ones run well on a Mac or iPhone

Leaderboard speed is measured on data-centre GPUs. On Apple devices, what matters is whether a model has a good Core ML or MLX conversion and fits in memory next to everything else.

• Parakeet TDT v2/v3 are the practical sweet spot: 600M parameters, Core ML conversions that run on the Neural Engine, long recordings in one pass. See running Parakeet on Apple silicon.
• Whisper has the most mature ports and the widest languages, at a cost in speed. See running Whisper locally on a Mac.
• Apple's SpeechAnalyzer, built into iOS 26 and macOS 26, needs no download and works offline for the languages Apple supports.
• Qwen3-ASR 0.6B, Granite 5.0 TurboCTC and Parakeet Ultra are the candidates to watch for on-device use; their size fits, and conversions are appearing.
• 2B-plus models with LLM decoders (Canary-Qwen, Granite 4.1, Cohere, VibeVoice) are more demanding and better suited to a Mac with plenty of memory than a phone.

Frequently asked questions

What is the most accurate open-source speech recognition model in 2026?

On the Hugging Face Open ASR Leaderboard for English (September 2026), Alibaba's Qwen3-ASR 1.7B leads the open models with a 4.31% average word error rate, followed by NVIDIA Canary-Qwen 2.5B at 4.43%. Cohere Transcribe leads the open models on long-form audio.

Is Whisper still the best open speech model?

No, not for English accuracy. More than a dozen open models now beat Whisper large-v3 on the Open ASR Leaderboard, including smaller and much faster ones like Parakeet TDT. Whisper is still valuable for its roughly 99 languages, translation to English, and mature tooling.

Which open speech model is fastest?

IBM's Granite Speech 5.0 TurboCTC, at nearly 13,000 times real time on the leaderboard's GPU, followed by NVIDIA's Parakeet TDT models at about 6,000. Both are also more accurate than Whisper large-v3.

Which open model supports the most languages?

Meta's Omnilingual ASR covers more than 1,600 languages. Among models focused on accuracy, Qwen3-ASR covers 52 languages and dialects, Parakeet TDT v3 and Canary-1B-v2 cover 25 European languages, and Whisper covers about 99.

Can I use these models commercially?

Most can. Qwen3-ASR, Granite, Cohere Transcribe and Voxtral use Apache 2.0; Parakeet and Canary use CC-BY-4.0, which requires attribution; Whisper and VibeVoice use MIT. Some variants, such as Granite 5.0 TurboCTC-nc, are non-commercial. Check each model card.

Sources

• Hugging Face Open ASR Leaderboard
• Open ASR Leaderboard results dataset
• Qwen3-ASR model card and technical report, arXiv:2601.21337
• Parakeet TDT 0.6B v3 model card
• Parakeet TDT 0.6B v2 model card
• "Canary-1B-v2 & Parakeet-TDT-0.6B-v3", arXiv:2509.14128
• Canary-Qwen 2.5B model card
• Nemotron 3.5 ASR streaming model card
• Parakeet Ultra model card (moondream)
• Granite Speech 5.0 TurboCTC announcement
• Granite Speech 4.1 2B model card
• Cohere Transcribe model card
• Mistral: Voxtral Transcribe 2
• VibeVoice-ASR, arXiv:2601.18184
• MOSS-Transcribe-Diarize model card
• Omnilingual ASR, arXiv:2511.09690
• Kyutai STT 2.6B model card
• OpenAI API changelog (Whisper deprecation)
• FluidAudio releases