Google has announced an update that completely resets usage quota counters for both free and paid tier developers. This move coincides with the deployment of a refined version of the Gemini 3.5 Flash model within the Antigravity platform, aimed at resolving recent output quality and performance issues.
The update seeks to correct a specific “blind spot” identified in a previous “low-effort” variant of the model. That variant was originally designed to optimize token efficiency and prevent the artificial intelligence from expending excessive computational resources on simple user requests. While the low-effort version successfully reduced overall token consumption by approximately 45% compared to the standard model, developers reported a sharp decline in output quality and structural consistency when handling analytical tasks of higher complexity.
Varun Mohan, a director at Google DeepMind who oversees the Antigravity project, stated that the updated model delivers superior performance when handling difficult and complex reasoning challenges. Google promises that the refined system offers greater robustness and stability, particularly when managing heavy computational workloads such as software programming tasks.
However, the company clarified that these distinct processing categories based on effort levels (Low, Medium, and High) are unique to the Antigravity development environment. There are currently no plans to introduce these options as manual selectors within the consumer-facing Gemini application.
Google indicated that the full reset of rate limits was implemented as a courtesy to allow developers to test the capabilities of the revised model immediately. The company is actively monitoring user feedback to address ongoing community requests. A primary focus in their suggestion channels is the implementation of a visual usage tracking bar, which would inform developers exactly how much of their allocated weekly quota has been consumed.
Filed in . Read more about AI (Artificial Intelligence), Gemini and Google.