Google has introduced a new artificial intelligence model variant called Gemini 3.5 Flash Low to help users of its Antigravity coding platform reduce token consumption and avoid usage limits.
The announcement comes after Google recently shifted Gemini plans from a message based system to a compute based usage model that measures token consumption.
Following the rollout, many users reportedly complained about reaching rate limits too quickly while using Antigravity for coding and development tasks.
In response, Google had already increased Antigravity rate limits twice, but users continued reporting issues related to token exhaustion.
Explaining the latest update, Varun Mohan said the new Gemini 3.5 Flash Low model was specifically designed to optimise token usage for simpler tasks without significantly affecting performance.
According to Mohan, the Low variant generates nearly 45 percent fewer output tokens compared to the standard Gemini 3.5 Flash model, now renamed Gemini 3.5 Flash Medium.
Google has also introduced another version called Gemini 3.5 Flash High, which is intended for more complex software engineering and coding operations.
The company claims that despite lower token consumption, the Low variant can outperform the High version in several software engineering benchmarks.
Alongside the update, Google has temporarily reset Gemini quotas across all subscription plans, including free tier accounts, allowing users additional usage capacity for ongoing projects.
Users have also raised concerns regarding image generation limits on the platform.
One user pointed out that they could generate significantly more AI images on competing platforms than on Antigravity despite subscribing to Google’s Ultra plan.
Responding to the criticism, Mohan acknowledged that the current image generation limit was too low and indicated that Google may consider increasing the cap in the future.
The latest changes reflect Google’s efforts to improve user experience and manage growing demand for AI powered coding tools amid rising competition in the generative AI sector.
