I really do He liked Antigravity, which Google announced last year. The reason was that it had a generous tariff limit and was quite functional based on its electronic design. Same foundation as VS Code. You can do a lot in Antigravity even using the Pro models, but now, when I returned to Gemini last week, things were very different. Now the tariff limits are a mess. If you are not a Pro member, the platform is not even usable. The free level is a trial that doesn’t even take half an hour if you try to do anything serious. I shouldn’t complain that I don’t get enough rate limits for free, but when you look at something like Codex, which has generous rate limits in both the Codex app and the VS Code extension, you start to question why you shouldn’t shut down Antigravity for good.
Antigravity’s free level is terrifying
You can do almost nothing using the free level
When Google first launched Antigravity, the developer community really thought it was potential to be a serious Cursor alternative. The combination of Gemini models, agent workflow, and VS Code-style Electron foundation made it surprisingly powerful, especially since the early access period came with relatively generous usage limits. You can actually sit and work with it for hours without constantly thinking about quotas. That version of Antigravity is gone.
My biggest complaint about Antigravity today is the free tier, which can’t be reliably used for real-world development work. The most common frustration is the 7-day lockout. There were times when I was offline for an entire week after only 20-30 minutes of active use. Once the system decides your weekly bucket is exhausted, Antigravity stops responding to requests until the cooldown is reset. This completely destroys the continuity of the workflow.
What makes matters worse is how inconsistent the quota system is. On Google’s own developer forums, users complain that the user interface can still show moments of “100% remaining capacity” before a single request causes a multi-day block.
Google also doesn’t disclose quota usage in a particularly understandable way. Instead of using a simpler token-type system like competitors, Antigravity relies on what it calls “computational effort.” Complex tasks like architecture planning, multi-file refactoring, or analysis of a larger codebase can consume a large portion of your weekly bill almost immediately, even if the visible interaction itself seems small.
The pro level isn’t that good either
You still quickly hit the rate limit
The Pro story is better than the free tier, but it’s still not impressive when you compare it to what other tools are doing right now. Google AI Pro costs $19.99 per month in the US, while Google AI Ultra costs $249.99 per month. Antigravity access is included in plans with higher rate limits in the agent model, but the official wording still stops short of anything that sounds unlimited or particularly generous.
Google says that Pro gets a higher quota refresh every five hours until the weekly limit is reached, while Ultra gets the highest rate limits. But even Pro users are seeing cooldowns or longer wait times of between 6 and 10 days on individual models, and Google says the weekly limits apply to all models and the Ultra is exempt from those limits.
When you compare these to what OpenAI offers with Codex, you start to see very little reason to stick with Antigravity, even though they are two different products. OpenAI’s Codex Pro starts at $100 per month and offers 5x or 20x higher rate limits than Plus with a temporary 2x usage promotion at $100. Codex is also included in ChatGPT Free and Go for a limited time, and higher limits apply in the app, CLI, IDE, and cloud.
Claude is not as easy a comparison as it used to be. Anthropic said on May 6, 2026, it doubled Claude Code’s five-hour rate limits for its Pro, Max, Team and seat-based Enterprise plans and eliminated peak-hour reductions for Pro and Max.
The rate cap problem is real
We’re at a point where AI models, coding agents, and new-age IDEs like Antigravity and Cursor are becoming really useful. They allow you to do more in less time, but they are also very expensive. You are already aware of the prices for the Pro and Ultra levels of tools and models such as Claude, Claude Code, Codex and similar solutions.
It’s the only practical long-term solution I see to use local models as much as possible. You can use on-premises models for tasks that don’t require massive computing, then switch to powerful cloud models for work that isn’t possible with a single on-premises model.
