Google Unveils Native Gemini Mac App, Flash TTS Voice, and TurboQuant Compression for

Google said the Gemini app now supports “TurboQuant” compression, a 3.5‑bit quantization technique that shrinks large‑language‑model key‑value caches by up to six times while preserving accuracy, according to an InfoQ report. The company claims the method requires no retraining and lets developers run longer context windows on modest hardware, a claim backed by early community benchmarks that show “significant efficiency gains” (InfoQ).

The Mac app also adds a new “Flash TTS” voice engine, Gemini 3.1 Flash TTS, which introduces granular audio tags for fine‑grained control over vocal style, pacing and delivery in more than 70 languages. Google’s blog notes the model produces “more natural” speech than prior versions and that every output is water‑marked with SynthID to curb misinformation (Google Blog). Developers can test the model in Google AI Studio, Vertex AI and Google Vids, and fine‑tune voices for consistent use across applications.

Gemini for macOS can be summoned from any window with Option + Space, while Option + Shift + Space opens a full‑screen chat. The app lives in the Dock and Menu Bar, and any shared window can be analyzed by Gemini, which reads full pages after receiving Accessibility permission (MacRumors). Built‑in tools include Nano Banana for image generation and Veo for screen analysis, allowing users to ask contextual questions without leaving their workflow.

Google positioned the native Mac client as the last of the three major AI services to receive a dedicated desktop app, noting that OpenAI and Anthropic already offer similar macOS experiences (MacRumors). By integrating Flash TTS and TurboQuant into the same ecosystem, Google aims to give developers a unified platform for multimodal generation, high‑quality speech and faster inference on lower‑end machines.