Google Gemini

Google Gemini is Google's consumer AI assistant, reachable at gemini.google.com, as a mobile app (Android/iOS), and as an overlay that can replace Google Assistant on Android. It descends from Bard, which Google unified under the Gemini brand on 8 February 2024. The underlying models have moved fast: Gemini 3 Pro launched 18 November 2025, Gemini 3.1 Pro followed in preview on 19 February 2026 (Google reports an ARC-AGI-2 score of 77.1%, more than double Gemini 3 Pro's, and 80.6% on SWE-Bench Verified — vendor-reported benchmark claims, not independently verified here), and by October 2026 the app and its Google AI subscriptions reference Gemini 3.8 Flash and a new multimodal "Gemini Omni" model introduced at Google I/O in May 2026. For a buyer this matters less as a spec sheet and more as a pattern: expect the specific model names behind each plan to be stale again within a few months.

As read today from Google's current Gemini-branded pricing page (gemini.google/subscriptions, Germany-priced, EUR): Free (€0/month, standard usage limits, Gemini 3.5 Flash-Lite); Google AI Plus (€4.99/month, 2x the free usage limits, access to Gemini 3.6 Flash and 3.1 Pro, 400GB storage, 200 Google Flow video-generation credits); Google AI Pro (€21.99/month, 4x the free limits, 1,000 Flow credits, 5TB storage, YouTube Premium Lite and Google Home Premium included); and Google AI Ultra, which starts at €99.99/month for 5x Pro's limits or €219.99/month for 20x Pro's limits, with 20TB+ storage and exclusive access to Deep Think reasoning mode and Gemini Spark. Separately, Google's general Google One plans page (one.google.com/about/plans), fetched the same day, lists "Google AI Plus" at €9.99/month with 2TB of storage — a different price and a 5x larger storage allowance than the Gemini-branded page quotes for the same-named plan. Both are Google's own live pages as of 12 October 2026; which one a given signup flow actually charges is not established here, so check the number on your own checkout page before subscribing.

Beyond the usage-limit multipliers, the free tier is capped at a 32,000-token context window; Plus raises that to 128,000 tokens; Pro and Ultra both jump to 1 million tokens, enough to hold roughly an hour of video or tens of thousands of lines of code in one prompt, per Google's own help documentation. Deep Research, Canvas, Gems, and Nano Banana image editing are available on every tier including Free; what scales with price is how many Deep Research reports and how much image/video generation you get before hitting a limit, and whether you get the Pro model at all (Free and Plus never get full access to Gemini 3 Pro in Google Search's AI Mode; Pro and Ultra do). Ultra alone unlocks "Deep Think" extended reasoning and early access to experimental features.

Per Google's Gemini Apps Privacy Notice (last updated 29 June 2026, hub updated 24 September 2026), Gemini retains your chat activity for 18 months by default, adjustable to 3 months, 36 months, or indefinitely, and you can delete chats manually at any time. However, a subset of chats is reviewed by human reviewers (including trained contractors) to improve the models, and Google is explicit that this reviewed data — along with associated metadata like device type, location, and feedback — is retained for up to three years regardless of your delete or auto-delete settings. The notice also warns against entering confidential information you wouldn't want a reviewer to see. On Google Play, Gemini's own Data Safety declaration states no data is shared with third parties, but that the app may collect location, personal info, and 11 other data categories, with data encrypted in transit and deletion requestable.

Gemini is installed 1B+ times on Android with a 4.5-star average across 48.6 million Play Store reviews, and rates 4.7 stars across 2.3 million App Store ratings (#3 on the US Productivity chart at the time of this check). Independent press coverage (via Wikipedia's sourced history) documents one serious incident: in February 2024, Gemini's image generator was found producing historically inaccurate depictions of people (e.g., racially diverse Nazi soldiers and Founding Fathers) and refusing to generate images of white people; Google paused people-image generation entirely, the controversy knocked 4.4% off Google's stock price, and the feature didn't fully return until late August 2024 on a new model. Day-to-day user reviews on both app stores are more mixed than the star averages suggest — common complaints include server-side usage throttling on paid plans and, as of September 2026, at least one visible review noting the app was "stuck" on Gemini 3.1 Pro since February 2026 while competitors shipped newer models faster. On iOS specifically, Google's own app description notes Gemini does not support native device actions like setting alarms or sending SMS, a gap that exists on Android where Gemini can replace Assistant outright.

Gemini is not API-less. Google publishes the Gemini API at ai.google.dev/gemini-api/docs, with a free-tier API key obtainable through Google AI Studio, SDKs for Python/JavaScript/Java/Go, and a documented model lineup (Gemini 3.8 Flash, 3.5 Flash-Lite, 3.1 Pro, Nano Banana for images, Gemini Omni Flash for video) separate from the consumer app's own model access. This is a distinct product from the free gemini.google.com chat app: the API is billed separately and is what third-party developers actually integrate against.

Gemini is a closed, proprietary hosted service. There is no open-source repository for the models or the consumer app; nothing here is self-hostable. The only open-source-adjacent artifact is Gemini CLI, a separate open-source terminal agent Google released in June 2025, which is not the same product as the Gemini app being reviewed here.