Cohere Transcribe Leads New Hugging Face Far-Field ASR Benchmark

Cohere

Jun 13, 2026

Cohere Transcribe ranked first on the new Hugging Face Far-Field ASR benchmark with a 17.9% word error rate. The model outperformed IBM Granite Speech by nearly 2 points and NVIDIA Parakeet by 3.6 points. This benchmark measures transcription accuracy in real-world, far-field audio environments like meeting rooms and contact centers, where the model ranked first across every metric.

Spaces treble-technologies/ffasr
WER
Average WER (%) · Lower is better
17.9 cohere-transcribe-03-2026
19.5 granite-speech-4.1-2b
21.5 parakeet-tdt-0.6b-v2
23.2 parakeet-tdt-0.6b-v3
24.5 granite-4.0-1b-speech
26.3 canary-1b-v2
30.3 parakeet-ctc-1.1b
31.1 canary-1b-flash
31.9 lite-whisper-large-v3-tur...
33.9 owsm_ctc_v4_1B — Cohere-transcribe achieves the lowest word error rate among compared speech models in this performance benchmark analysis.

View the full update on huggingface.co

Cohere

@cohere3d ago

Cohere Transcribe, our open-source speech recognition model, is #1 on the new @huggingface Far-Field ASR benchmark. https://t.co/biwjzZosr8

48524

View on X

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Keep reading

Cohere Releases Open Source Transcribe Model Outperforming Whisper on Accuracy

Cohere launched Transcribe, a 2-billion parameter open-source speech recognition model that currently holds the top spot on the HuggingFace Open ASR Leaderboard. By achieving a 5.42% word error rate, it provides a high-accuracy, high-throughput alternative for enterprise workflows that previously relied on larger or proprietary models.

Artificial Analysis Benchmarks Google's Gemma 4 12B Transcription at 8.8% WER

Artificial AnalysisJun 6

Artificial Analysis Benchmarks Google's Gemma 4 12B Transcription at 8.8% WER

Artificial Analysis benchmarked Google DeepMind's new open-weight Gemma 4 12B model for transcription, reporting an 8.8% Word Error Rate (WER). This places the model behind specialized open-weight transcription solutions, but it is available for local deployment alongside Google's new Eloquent dictation app.

Together AI Launches Unified Voice Agent Cloud With Full Pipeline Co-Location

Together AIMar 18

Together AI Launches Unified Voice Agent Cloud With Full Pipeline Co-Location

Together AI launched a unified platform for real-time voice agents with STT, LLM, and TTS co-located on one cloud. Most voice stacks route audio across separate vendors — Together keeps all three in the same cluster, hitting latency under 700ms.

Cognition's Devin Integrates Claude Fable 5, Leads Real-World Engineering Benchmark

Cognition4d ago

Cognition's Devin Integrates Claude Fable 5, Leads Real-World Engineering Benchmark

Cognition has made Anthropic's Claude Fable 5 model available within its Devin AI software engineer across Cloud, Desktop, and CLI. This integration positions Fable 5 as the top performer on Cognition's FrontierCode benchmark, highlighting its advanced capabilities for production-grade code quality and mergeability in autonomous engineering tasks.