Google Antigravity director Varun Mohan announced on May 25 that the platform is adding Gemini 3.5 Flash (Low) as a new model tier to reduce token consumption on routine tasks. According to Mohan, internal testing shows the Low tier generates roughly 45% fewer tokens than Gemini 3.5 Flash (Medium) while still outperforming Gemini 3 Flash (High) on software engineering benchmarks. Mohan also confirmed that Gemini token quotas have been reset across all paid plans — later clarified to include free plans as well — to give developers fresh headroom heading into the week.
The move follows user complaints that Antigravity 2.0, which launched at Google I/O on May 19, was consuming excessive tokens even on straightforward edits. Mohan acknowledged the team had a “blind spot” in measuring token usage for simpler workloads, having optimized primarily for complex task performance. He clarified that the Low tier works by adjusting the model’s effort level during inference — a behavior the model learns in training — rather than truncating system prompts or compacting context, so overall experience quality is not intended to degrade. The team said it plans to improve cost measurement for varied task types going forward.