When OpenAI released Whisper in 2022, it changed what developers expected from on-device speech recognition, trained on 680,000 hours of multilingual audio and capable of transcribing in 99 languages. The article describes what happened when the authors ran their on-device model, Speechmatics On-Device, against the strongest Whisper apps using the same hardware and the same 20 minutes of clean audio. On Windows with RTX 4050, Speechmatics On-Device reaches 25.3 s/s using 1.7 GB total memory, while the closest Whisper build hits 22.1 s/s but uses 3.2 GB. On macOS M1, Speechmatics On-Device on ANE/GPU reaches 47.2 s/s at 1.1 GB, while WhisperKit CLI runs at 11.7 s/s, a 4x speed difference on Apple Silicon. One caveat: WhisperKit CLI uses only 0.46 GB on M1, less than half the footprint. The article notes that memory determines whether a model is actually deployable.