What was announced

Google rolled out a press-and-hold voice interaction layer for the Gemini macOS app, letting users speak to Gemini from any window on their desktop by long-pressing the Fn key . The update delivers two distinct modes.

The two modes and how they work

1. Intelligent Dictation (default, always-on)

• Long-press the Fn key in any app and speak naturally .
• Gemini transcribes your words into clean, polished text that automatically formats itself — handling punctuation, capitalization, and even removing filler words ("um," "ah," and mid-sentence corrections) .
• Output appears inline in whatever app or text field you're using. Designed as a smarter replacement for macOS's built-in dictation .

2. Screen Awareness (opt-in)

• You must manually enable this feature (opt-in only) .
• Once enabled, Gemini can "see" what's on your screen through screen-sharing permission .
• You can issue voice commands that reference on-screen content — e.g., "summarize this document" or "edit that paragraph" — and Gemini acts on the visible content .
• The Verge describes this as letting Gemini "perform tasks within your open windows" .

Privacy trade-off

• Intelligent Dictation processes audio in Google's cloud; standard Gemini privacy policies apply (chats stored by default for 18 months, with options for 3/36 months or auto-delete, and human review possible on opted-in chats) .
• Screen Awareness is the bigger privacy consideration: it requires granting Gemini screen-recording permissions. Because it's opt-in, nothing is shared unless you explicitly toggle it on . Once enabled, Google can see the contents of your active windows, which raises the same class of concerns as any AI screen-reading tool (accidental sharing of sensitive info, credential visibility, etc.). Google says it checks with you before taking major actions, but the screen data itself is processed on Google's servers .

Comparison: Gemini vs ChatGPT Desktop vs Apple Intelligence

Key differentiators: ChatGPT Desktop's GPT-Live-1 is a true conversational voice mode — it can listen and speak at the same time, handle interruptions, and direct AI agents in real time . Gemini's approach is more task-oriented: dictation plus screen-aware commands rather than fluid back-and-forth chat. Apple Intelligence is deeply on-device and system-integrated but lacks the generative depth and cross-app autonomy of Gemini or ChatGPT.

Availability details

New Voice Control (July 29, 2026)

• Language: English initially; no official list of additional languages has been published .
• App version: Latest Gemini for macOS app (rolling update with no specific version number released) .
• Global rollout: Rolling out globally starting July 29, but may take days to reach all regions .
• Age restriction: 18+ (standard Gemini age policy).
• Subscription: Free for Intelligent Dictation; Screen Awareness is also free at launch — no Google AI Ultra or paid tier required for either voice feature .

Gemini Spark (arrived June 30, 2026)

• What it is: A 24/7 persistent AI agent that runs in Google Cloud, even when your devices are off. It automates tasks across Gmail, Docs, Calendar, files, and third-party services using Schedules and Skills .
• Language: English only at launch .
• App version: Requires the latest Gemini for macOS app (beta); Spark also works on Android, iOS, and web .
• Global rollout: Limited beta — US only as of June 30 . An India rollout was separately reported on July 29 .
• Age restriction: 18+.
• Subscription: Google AI Ultra subscribers only (paid tier). Not available on free Gemini .
• Key difference from voice control: Spark is a background agent (proactive, 24/7 automation), while voice control is a real-time input method (reactive, on-demand). Spark runs autonomously in the cloud; voice control requires you to be at your Mac and actively speaking. Spark costs money; voice control is free.

What was announced

Google rolled out a press-and-hold voice interaction layer for the Gemini macOS app, letting users speak to Gemini from any window on their desktop by long-pressing the Fn key . The update delivers two distinct modes.

The two modes and how they work

1. Intelligent Dictation (default, always-on)

• Long-press the Fn key in any app and speak naturally .
• Gemini transcribes your words into clean, polished text that automatically formats itself — handling punctuation, capitalization, and even removing filler words ("um," "ah," and mid-sentence corrections) .
• Output appears inline in whatever app or text field you're using. Designed as a smarter replacement for macOS's built-in dictation .

2. Screen Awareness (opt-in)

• You must manually enable this feature (opt-in only) .
• Once enabled, Gemini can "see" what's on your screen through screen-sharing permission .
• You can issue voice commands that reference on-screen content — e.g., "summarize this document" or "edit that paragraph" — and Gemini acts on the visible content .
• The Verge describes this as letting Gemini "perform tasks within your open windows" .

Privacy trade-off

• Intelligent Dictation processes audio in Google's cloud; standard Gemini privacy policies apply (chats stored by default for 18 months, with options for 3/36 months or auto-delete, and human review possible on opted-in chats) .
• Screen Awareness is the bigger privacy consideration: it requires granting Gemini screen-recording permissions. Because it's opt-in, nothing is shared unless you explicitly toggle it on . Once enabled, Google can see the contents of your active windows, which raises the same class of concerns as any AI screen-reading tool (accidental sharing of sensitive info, credential visibility, etc.). Google says it checks with you before taking major actions, but the screen data itself is processed on Google's servers .

Comparison: Gemini vs ChatGPT Desktop vs Apple Intelligence

Key differentiators: ChatGPT Desktop's GPT-Live-1 is a true conversational voice mode — it can listen and speak at the same time, handle interruptions, and direct AI agents in real time . Gemini's approach is more task-oriented: dictation plus screen-aware commands rather than fluid back-and-forth chat. Apple Intelligence is deeply on-device and system-integrated but lacks the generative depth and cross-app autonomy of Gemini or ChatGPT.

Availability details

New Voice Control (July 29, 2026)

• Language: English initially; no official list of additional languages has been published .
• App version: Latest Gemini for macOS app (rolling update with no specific version number released) .
• Global rollout: Rolling out globally starting July 29, but may take days to reach all regions .
• Age restriction: 18+ (standard Gemini age policy).
• Subscription: Free for Intelligent Dictation; Screen Awareness is also free at launch — no Google AI Ultra or paid tier required for either voice feature .

Gemini Spark (arrived June 30, 2026)

• What it is: A 24/7 persistent AI agent that runs in Google Cloud, even when your devices are off. It automates tasks across Gmail, Docs, Calendar, files, and third-party services using Schedules and Skills .
• Language: English only at launch .
• App version: Requires the latest Gemini for macOS app (beta); Spark also works on Android, iOS, and web .
• Global rollout: Limited beta — US only as of June 30 . An India rollout was separately reported on July 29 .
• Age restriction: 18+.
• Subscription: Google AI Ultra subscribers only (paid tier). Not available on free Gemini .
• Key difference from voice control: Spark is a background agent (proactive, 24/7 automation), while voice control is a real-time input method (reactive, on-demand). Spark runs autonomously in the cloud; voice control requires you to be at your Mac and actively speaking. Spark costs money; voice control is free.