Nothing introduces an AI-powered dictation tool

The announcement leads with breadth (100-plus languages) but says nothing about depth: no word-error rates, no comparison against existing on-device alternatives like Apple's dictation or Android's built-in speech recognition, and no clarity on whether low-resource languages get the same quality as high-resource ones.

The on-device speech push sits in a broader moment where AI audio capabilities are advancing quickly at the model level. Google DeepMind's Gemini 3.1 Flash TTS release from mid-April showed how much expressive control is now possible in cloud-side speech synthesis, which makes Nothing's local-first framing a deliberate counter-positioning rather than a pure technical necessity. The question is whether Nothing's hardware can actually deliver competitive quality offline, or whether 'on-device' is doing marketing work that the underlying model performance doesn't yet support. The related coverage here is largely about generative speech output rather than transcription input, so the direct technical comparison is limited.

Watch whether independent testers find meaningful accuracy gaps between Nothing Dictation and on-device Apple or Google alternatives on the same language set within the next 60 days. If the gaps are large, the 100-language claim is a feature count, not a quality commitment.