## Google Responds To Early Antigravity Friction Antigravity is part of Google's broader push into AI-assisted software development, where agents can plan, edit, test, and iterate across a codebase rather than simply autocomplete a line of code. That makes usage limits more sensitive than they are in a conventional chatbot. A coding agent can burn through model calls during planning, terminal work, test repair, and repeated file edits. For paid users, a quota that feels generous in ordinary chat can feel tight inside a development loop. Once an agent becomes part of the workflow, the interruption is not just an inconvenience. It can stop a task midstream, forcing users to wait, switch models, or move to a competing tool. That is why Google's quick response matters. By tripling rate limits and resetting weekly quotas, the company is trying to protect momentum with early adopters while the product is still forming habits. The message is not only that Antigravity is improving. It is that Google is watching user pain closely and is willing to adjust the product's economics quickly. ## Gemini Performance Raises Expectations The rate-limit change lands just days after Google highlighted Gemini 3.5 Flash as a strong performer for agentic coding tasks. According to the source packet, the model reached 76.2% on Terminal-bench and 55.1% on SWE-Bench Pro, benchmarks often used to evaluate how well AI systems handle command-line and software engineering tasks. Those numbers help explain why users may have pushed the platform hard immediately. A capable coding model invites heavier usage because developers start asking it to do more than answer questions. They ask it to investigate bugs, modify files, run tests, compare approaches, and recover from failures. That shift changes the pricing and quota problem. The best model on paper can still feel limited if users spend too much time managing caps. In the agentic coding category, product trust depends on a mix of model quality, reliability, speed, context handling, and the freedom to keep working through a real engineering session. ## Usage Economics Are Becoming A Product Feature Google is not alone in facing this tension. AI coding tools across the market are trying to balance expensive inference costs against users who increasingly expect agents to run longer, inspect more context, and complete more complicated work. The more autonomous these systems become, the more visible rate limits become to the user. That creates a strategic challenge for platform teams. Restrictive limits can protect compute budgets, but they can also make a tool feel unreliable at exactly the moment it starts proving useful. More generous limits can accelerate adoption, but they require confidence that model serving costs, subscriptions, and infrastructure can support the usage. For Antigravity, the decision to triple Gemini limits suggests Google wants to remove a blocker before it defines the product's early reputation. The company still has to clarify how permanent limits, compute-based caps, and tier differences will work over time. But for now, the signal is simple: Google would rather give paid Antigravity users more room to build than let quota frustration become the story. --- *Originally published on [Daily AI Mail](https://dailyaimail.news)*An update: we're 3xing the rate limits for Gemini models across all paid tiers in Antigravity and resetting everyone's Gemini quota for the week.
— Varun Mohan (@_mohansolo) May 21, 2026
We understand some people hit their rate limits quickly and wanted to respond fast. Lots more to come and enjoy building!