The End of the Whisper Monoculture
Open speech recognition stopped being a Whisper monoculture sometime in the last twelve months. In March 2026, Cohere Transcribe, a 2B Apache 2.0 model, took the top of the Open ASR Leaderboard at 5.42% average word error rate. Five weeks later, IBM's Granite Speech 4.1 2B edged it at 5.33%. Since then, ARK-ASR-3B and MOSS-Transcribe-preview-2B have posted lower numbers still.
The top of the leaderboard is now separated by less than one WER point. That has a specific consequence for anyone choosing a model: license, language coverage, streaming support, and cost per audio-hour matter more than marginal WER differences.
A Warning About Benchmarks
The Open ASR Leaderboard average is not a single fixed quantity. Cohere's 5.42% is an average across eight English test sets including TED-LIUM. ARK-ASR-3B's 5.04% is an average across seven sets, with TED-LIUM explicitly excluded. TED-LIUM is one of the easier sets, so dropping it raises the average. Recompute Cohere's scores over the same seven sets and it lands at 5.84, not 5.42.
MOSS-Transcribe was fine-tuned with reinforcement learning on the Open ASR Leaderboard training splits. That is disclosed, but it means the score measures the benchmark rather than the capability. Appen's private evaluation sets, covering Australian, Canadian, Indian, and American accents in scripted and conversational conditions, reorder the rankings when toggled on.
Use the leaderboard to build a shortlist. Do not use it to pick a winner.
The Production-Ready Contenders
Cohere Transcribe (2B, Apache 2.0, 14 languages) is the model that actually shipped into production. It has been downloaded over 620,000 times in the past month and has runtime support across ONNX Runtime, vLLM, mlx for Apple Silicon, a Rust port, and a WebGPU build. It is a Conformer encoder with a lightweight Transformer decoder, trained from scratch. Cohere also ran human preference evaluation, where trained annotators scored transcripts for meaning preservation, hallucination, and named entities: a 61% average win rate, 78% against IBM Granite 4.0 1B Speech and 64% against Whisper large-v3.
The limitations are honest: no automatic language detection, no timestamps, no diarization, and the model is eager to transcribe silence. Cohere recommends prepending a VAD or noise gate. The repo is gated behind a contact-information agreement despite the Apache 2.0 license.
IBM Granite Speech 4.1 2B (Apache 2.0) is the better pick if you need capability over a lower number. Six languages for ASR plus bidirectional speech translation, keyword-list biasing for names and jargon, punctuation and truecasing including German noun capitalization. Trained on 174,000 hours. RTFx 231.29. IBM also ships Granite Speech 4.1 8B, which adds speaker-attributed ASR and word-level timestamps.
Canary-Qwen-2.5B (CC-BY-4.0, English) pairs a FastConformer encoder with a Qwen3-1.7B decoder and runs in two modes: pure transcription, or LLM mode where the decoder summarizes and answers questions about the transcript. 5.63% WER at RTFx 418. Note that AMI was oversampled to roughly 15% of training data, which biases output toward verbatim disfluency-preserving transcripts. A feature for legal work, a nuisance for meeting notes.
Qwen3-ASR (Apache 2.0) covers 52 languages and dialects, 30 languages plus 22 Chinese dialects, at 5.76%. It ships with a full inference toolkit and a separate forced-alignment model for timestamps in 11 languages. For anything touching Mandarin or Chinese regional speech, this is the obvious starting point.
Throughput: The Real Decider
Accuracy across the top of the field varies by about one WER point. Throughput varies by more than an order of magnitude, which means throughput usually decides the invoice.
Parakeet TDT 0.6B v3 (CC-BY-4.0) is the throughput leader among multilingual open models at RTFx 3332.74 across 25 European languages with automatic language ID, handling up to 24 minutes in a single pass on an A100 80GB. It costs 6.32% WER, roughly one point more than Granite 4.1 2B for roughly fourteen times the audio per GPU-second.
ARK-ASR-3B is the more interesting engineering result. It is non-autoregressive: it edits a CTC hypothesis in a single forward pass using a bidirectional LLM, reaching RTFx ~1820 on one H100 at batch size 128. It gives up Japanese, speech translation, and keyword biasing to get there.
Streaming Models
A batch WER is the wrong test for a streaming model, and the leaderboard scores them anyway. Voxtral Realtime sits at 7.68% and Kyutai STT 2.6B at 6.40%, both below Whisper, and neither number tells you anything useful about their intended use.
Voxtral Realtime (Apache 2.0, 13 languages) is a 3.4B language model with a 970M causal audio encoder trained from scratch, with sliding-window attention for effectively unbounded streaming. Transcription delay is configurable in 80ms steps from 80ms to 1200ms. Mistral recommends 480ms as the sweet spot and reports that at that setting it matches leading offline open models. It runs on a single 16GB GPU.
Kyutai STT (CC-BY-4.0) comes in two shapes: a ~1B English/French model with a 0.5s delay and a built-in semantic voice activity detector, and a 2.6B English-only model with a 2.5s delay. For voice agents, the semantic VAD matters more than transcription delay. It predicts when the speaker has actually finished, which governs perceived turn-taking latency. An H100 serves 400 concurrent streams in real time.
The Long Tail
Omnilingual ASR (Apache 2.0) is not competing on WER. It covers 1,600+ languages natively and extends to 5,400+ through zero-shot in-context learning, built on a wav2vec 2.0 encoder scaled to 7B and pre-trained on about 4.3M hours. The 7B LLM-ASR variant achieves character error rate below 10% on 78% of supported languages, including 500+ never previously served by any ASR system.
Whisper large-v3 (1.55B, MIT, 99 languages) has been overtaken on accuracy by roughly ten open models, and it remains the correct default for a large class of projects. MIT is the least encumbered license in the field. The runtime ecosystem, whisper.cpp, faster-whisper, WhisperX, has no equivalent among newer releases. If your requirement is "some language, some hardware, no license lawyer," it is still the answer.
Interfaze Diffusion ASR from YC startup Interfaze is the most architecturally creative release of the year. It produces transcripts by parallel diffusion denoising over a 256-token canvas in 8 to 16 steps, so decoding cost does not grow with transcript length. Only ~42M parameters were trained on top of a frozen 26B DiffusionGemma and a frozen whisper-small encoder. It reaches 6.6% WER on LibriSpeech test-clean at roughly 11 to 17x realtime. Nobody should deploy this. Everybody working on ASR should read it.
Canary-ASR 128k (Apache 2.0, 50+ languages) solves the problem most roundups ignore: it emits speaker labels, word timestamps, and transcript in one generation instead of chaining ASR to a separate diarization stack. 128k context, roughly 90 minutes of audio in one pass, RTF ~0.017 on an RTX 4090, with hotword biasing.
How to Choose
- If attribution is a blocker, CC-BY-4.0 removes Parakeet, Canary-Qwen, and Kyutai before you benchmark anything.
- Language limits are hard, not soft: Cohere's 14 languages, Granite's 6, and Canary's English-only are real boundaries. Cohere additionally has no language auto-detection.
- Streaming is architectural. No amount of tuning turns an offline encoder-decoder into a low-latency streaming model.
- Your audio matters more than the leaderboard. The spread between the top ten models on the public benchmark is under one point. The spread on your accented, noisy, domain-specific audio will be several times that.
- RTFx figures are measured at large batch sizes on datacenter hardware and do not transfer directly to your setup.
The useful summary of 2026 is not that any single model won. It is that a 2B open-weight model on a permissive license now beats what closed APIs were charging for eighteen months ago, and that the remaining decision is a procurement question rather than a research one.
Leaderboard positions are live and change frequently. All figures verified against primary sources on 23 July 2026.