Google Restricts Free Gemini Users to Flash-Lite Model Starting October 9
Google is restricting free Gemini users to its smallest model, Flash-Lite (3.5), removing access to the Flash (3.6) and Pro (3.1) models starting October 9, mynet.com reported. CEO Sundar Pichai stated that the standalone application surpassed one billion monthly users over the summer, with the vast majority of those users paying no fee. Under the updated support documentation, users without an active Google AI subscription will no longer have access to model switching in the application interface. DonanımHaber reported that the changes apply globally to all users without a paid subscription, following a shift toward processing power-based usage limits introduced by Google in May.
AI Plus Subscribers Lose Pro Access While Pricing Tiers Shift
Google AI Plus subscribers paying 4.99 dollars per month will retain access to Flash-Lite and Flash models but will lose access to the Pro model, according to mynet.com. DonanımHaber noted that AI Plus subscribers will receive an email detailing when the changes take effect for their accounts. Meanwhile, higher-tier plans maintain broader access: AI Pro and AI Ultra subscribers keep full access to Flash-Lite, Flash, and Pro models. As reported by mynet.com, 19.99 dollar monthly AI Pro subscribers gain access to Deep Think, a parallel reasoning feature previously restricted to plans priced between 99.99 and 199.99 dollars per month.
Three New Reasoning Levels Deploy to Gemini Application
Gemini is rolling out low, medium, and high thinking levels for each available model within the application, replacing the single extended thinking option for complex problem solving. These new settings will match options currently offered in Antigravity and Google AI Studio. While higher settings improve comprehensive response capabilities and thorough task completion, they consume a significantly larger share of the user’s available usage quota.

Gemini Gem Integration and Google Assistant Transition Continue
Google is updating its ecosystem alongside the model changes, as log.com.tr reported that Gemini’s custom “Gem” features will transition into “Skills” starting November 17, automatically migrating existing user configurations. In parallel mobile developments, cnnturk.com reported that Google has begun disabling Google Assistant on Android phones, tablets, Wear OS watches, and wireless headphones to complete its transition to the Gemini ecosystem. This process began September 3 following warning emails sent in August, resulting in the removal of the “return to Google Assistant” option on many devices. Users are advised to select Gemini as the default assistant in the Google Digital Assistants settings section. While Gemini handles complex logical requests, it still carries the risk of “hallucinations” and is not yet at the level of the old Assistant for offline voice commands.
Gemini 4 Argon
Google announced that the Gemini 4 Argon model will first be available to AI Ultra subscribers. However, the company has not yet clarified if Argon will be classified as a Pro model or as part of a new tier above Pro.
Frequently Asked Questions About Gemini Model Limits
Which models remain available for free Gemini users after October 9?
Users without a paid subscription are limited to the Flash-Lite (3.5) model and lose access to Flash and Pro models, according to mynet.com.
How much does the AI Pro subscription cost, and what does it include?
The AI Pro subscription costs 19.99 dollars per month and provides access to Flash-Lite, Flash, and Pro models alongside the Deep Think reasoning feature, as reported by DonanımHaber.
When will the new thinking levels be added to the Gemini application?
Low, medium, and high thinking levels are scheduled to arrive in the Gemini application during the month of October, according to Martin Cid Magazine.
What is happening to Gemini’s custom Gems?
Google announced that Gems will convert into Skills starting November 17, with automated migration handling user configurations during the transition, as reported by log.com.tr.
Worth a look