Google has recently updated its support page announcing that starting October 9, it will adjust access permissions and usage limits for Gemini models on personal accounts, comprehensively tightening model configurations for lower-tier users.
Under the new rules, unpaid free users will only be able to access the Gemini3.5-Flash-Lite version and will no longer have access to Gemini3.6 Flash and newer models; AI Plus subscribers at $4.99 per month will have their usage scope compressed to the lightweight Flash and Flash-Lite models, with no longer support for advanced reasoning models such as Gemini3.1Pro.
This change does not affect Google AI Pro and Ultra premium subscribers starting at $19.99 per month, who will continue to enjoy the Deep Think option and access to advanced models.
In addition to downgrading model tiers, Google has also introduced three workload (Effort) mode options: low, medium, and high; the system will deduct usage in real time based on the computational complexity invoked by a task, and higher workload modes will consume quota limits faster. Background information shows that Google has recently begun testing the next-generation Gemini4 "Argon" model in closed environments, and continues to iterate frequently on lightweight models such as Gemini3.8Flash.
Facing rising inference costs for frontier models and high-concurrency compute loads, leading AI companies are gradually ending the "high-specification free trial" phase. Through this tiered, fine-grained quota control, Google is tilting high-compute models toward high-value paying users, marking the shift of commercial large model applications from "user scale expansion" into a new phase of "refined operation for compute cost-effectiveness."