Apr 8

Local transcription using cohere-transcribe-03-2026

# Current situation I currently rely on Parakeet TDT 0.6B for local transcription. While it works well for everyday use, I've been looking for an option that offers both better transcription accuracy and higher throughput - especially for longer recordings or batch processing. # Proposed solution I'd like to see Cohere Transcribe 03-2026 added as an available model for local transcription. Model details and release announcement: https://huggingface.co/blog/CohereLabs/cohere-transcribe-03-… According to the official release, this model delivers state-of-the-art transcription quality while maintaining strong inference performance. The recommended way to run it locally is via vLLM, which SuperWhisper could integrate as the inference backend for this model. # Why this would be valuable - Improved transcription accuracy compared to smaller models like Parakeet TDT 0.6B - Better throughput for users who process longer or more frequent recordings - vLLM is a well-established, high-performance inference engine, making it a solid foundation for integration - Gives users a meaningful upgrade path without switching to a cloud-based solution

Pending