Google is expanding Antigravity's Neon cache pool and expects to be able to complete at least 20% more work with the same quota, and promises to reset the quota after the adjustment is completed. Director Varun Mohan said this morning that migration of large clusters would still take time. (Preliminary news: Google launches Gemini 4 Neon flagship model, which can be used with Pro subscription) (Background supplement: Google admitted its mistake after community bombardment: Antigravity Gemini rate limit increased 3 times, reset this week's quota) Varun Mohan, head of the Antigravity team of Google DeepMind, published a document at 2:38 a.m. on November 21, Taipei time, stating that Neon is being significantly expanded. Cache pool. He expects that after the improvement, users can complete at least 20% more work within the same quota; since the previous quota was consumed quickly, it will be reset after the adjustment is completed. Google recently launched Gemini 4 Neon, which focuses on long-running program development work that requires repeated calls to tools. It is also open to paid Pro and Ultra subscribers of Antigravity. Tech content creator Chubby (@kimmonismus) praised the model’s performance on X and said he has subscribed to Ultra. Longer intervals between model calls make caches easier to clear. Mohan explained in a 2:45 AM addendum that jobs with long intervals between model calls have particularly poor cache hits because the caches have been cleared. He expects that this expansion can improve such problems, and the affected work should be improved more significantly. Mohan estimates that significantly expanding the cache pool can increase Neon's hit rate by about 4%, thereby allowing at least 20% more work to be completed with the same amount of money. He also said that the team will continue to improve this indicator. The reset is scheduled after the migration is completed, and the team also clarified the issue of automatic compression. At 7:57 a.m., a user asked when the quota would be reset. Mohan replied that the reset would be carried out, but the team was doing a large-scale cluster migration and it would take some time. This continues the sequence he proposed previously: complete infrastructure adjustments first and then deal with quota resets. On the other hand, user David Shapiro complained about the credit consumption of long conversations and suspected that the platform did not automatically compress the context. Mohan responded at 9:50 AM that Antigravity already has automatic compression and the cache is being improved. The cache reuses the context, and the subscription quota is calculated based on the workload. Google's Gemini API document states that the implicit cache will operate automatically. When the request hits a common prompt prefix, processing costs can be saved; requests with similar prefixes are sent in a short time, which is easier to hit. The cache of Gemini 4 is automatically processed by the system and does not support users to manually create and manage cache objects. The program agent's subsequent work will use previous conversations, code snippets, and tool results. Caching reuses processed content; automatic compression reduces the context so that the model needs to process less content when it continues to work. The two have different functions. As for the Antigravity subscription quota, the official plan description states that the consumption is related to the work completed by the agent: simple tasks can handle more, while complex tasks consume more quota. This is how subscription usage is managed, and is different from API token-based billing. What this adjustment wants to improve is to allow the same share to support more practical work. Related reports Google launches Antigravity 2.0: Gemini 3.5 Flash engine, dynamic sub-agent, and Scheduled scheduling functions are in place Google launches Gemini 4 Hydrogen! 4 levels of thinking depth, focusing on daily programming and AI agents Google releases Gemini 4 Argon! The great evolution of spatial reasoning, driving the Apollo humanoid robot, API pricing released, Gemini Spark renamed "Task", open to 6 countries for free, Taiwan requires a subscription to use This article was first published in BlockTempo "Dongqu Dongzhu - the most influential blockchain news media".