Beating AI News Flash: Google's Gemini 4 Neon coding capabilities have received praise from many developers, but the accompanying coding tool Antigravity has faced concentrated complaints due to abnormal Token consumption.
In the early hours of November 21, Antigravity lead Varun Mohan admitted that Neon's cache hit rate is low, especially when the interval between two model calls is long, the cache is easily cleared. Google is significantly expanding the cache pool, expected to increase the hit rate by about 4%, allowing users to complete at least 20% more work within the same quota. User quotas will also be reset after completion.
Each time a coding agent calls the model, it needs to provide the model with previous conversations, code, and tool execution results. Caching allows the model to directly reuse already processed content without recalculating. Once the cache expires, hundreds of thousands of Tokens of historical content may be reprocessed, causing a significant increase in quota consumption.
