Cohere releases open source model that tops speech recognition benchmarks

Editor's take

Cohere has made an open-source speech recognition model available that, on the Hugging Face Open LLM Leaderboard, outperforms existing models like OpenAI's Whisper.

This release democratizes access to high-performing speech-to-text technology, potentially lowering barriers for developers and researchers working on applications ranging from accessibility tools to transcription services. The move challenges the dominance of proprietary models and fosters a more competitive ecosystem for AI-powered audio processing.

Future developments will reveal how quickly other organizations can integrate and improve upon this model, and whether Cohere’s strategy of open-sourcing a benchmark-leading model translates into broader adoption and influence in the LLM space. The performance of this model on real-world, noisy audio will be a key indicator of its practical impact.