Context
I subscribed to Google AI Pro 2 days ago. During this period, I have never hit my 5-hour rate limit. To the best of my recollection, I have used Claude Opus 4.6 on 2 moderately complex tasks:
-
A refactor of ~10 scripts, each less than 200 lines, and most less than 100.
-
Coming up with an implementation plan for a new feature, which was then executed through Jules. This feature involved minor edits to ~10 existing scripts (about 1-3 line edits each), and the creation of 3 new scripts, the largest of which was about 150 lines.
I have also used it to come up with a handful of other less demanding bug fixes and implementation plans, which were then executed by Claude Opus 4.5, Gemini 3 Pro or Flash. I am smart about managing my context window. I don’t let my chats get too long. I start new chats for new issues. I provide precise context. I provide prompts which are well-structured, detailed and narrowly-scoped .
Last night when I went to bed, my Antigravity Toolkit showed Claude usage as 20% used, with a reset timer of about 4 hours. This morning when I woke up it shows a usage of 60% used, with a reset timer of 5 days. Literally nothing has happened during that time, except that some algorithm on a Google server broke.
This Is False Advertising
The Google AI Pro subscription page I signed up on simply says:
Higher rate limits to agent model in Google Antigravity, our agentic development platform
which suggests, at minimum, that this tier is intended to be used by people who want to use Antigravity for real-world development tasks. If it wasn’t intended for this purpose, this tier would only include the more consumer facing AI options like Whisk, Flow, and usage in basic Google apps like Mail and Docs. The average person doesn’t need access to an agentic IDE. The only people who use such tools are developers. Ergo, this tier is aimed at developers.
It’s well known that the rate limits for model usage in Antigravity are completely opaque, and intentionally so. Even the tool I’m using to monitor mine is simply a best-guess based on local metadata and session headers. Google claims this is because it would be sooooo painfully difficult for them to show us what we’ve used, and what our limits are, due to the way usage is actually calculated by work done. But, considering how completely this issue has affected the entire developer community, I think the answer is much simpler: Google has set extremely low model usage limits for this tier, and wants us to believe otherwise. To be perfectly blunt, I would not be at all surprised to find out that the Pro and Free tiers share the exact same rate limits. I would even go so far as to say that I believe that the 5 hour refresh on the Pro tier is a straight-up lie. Maybe it used to be true, I’ve heard the rate limits used to be considerably better, but it is absolutely not true now that the 5 hour rolling reset works as advertised. From my experience, the misleading 5 hour reset is false advertising. It is consumer fraud.
To The Staff Who Read This
I know you guys are just doing your jobs. These kinds of decisions are made by executives that care more about money than they do about people, not by developers and customer service reps. But a lot of people, myself and the majority of the developer community using Antigravity included, are very angry about this situation, and justifiably so. We are paying a premium price for a product that has been gutted. We are paying a premium price to not receive what we thought we were buying. We are paying a premium price to try to do our real-life jobs that we use to buy food and pay our bills and support our families by using your product, and finding out that we simply can’t.
Please do not tell me to file a bug report. For one, this is not a bug, it is working as intended. For another, the message I see copy-pasted onto every forum post about this issue says to use the Feedback button in the top-right corner of the Antigravity UI, which simply doesn’t exist. There is a Report Issue link hidden under my account dropdown, but it’s not where you keep telling people it is, and it’s called something different than you keep telling people it is.
I expect to be told that you will escalate this issue to the appropriate people and nothing else, which, whatever, that’s fine and I suppose the best I can hope for in the short term. But please don’t pretend like we are doing something that you didn’t intend or expect us to do. We are simply trying to use the product as advertised.
What We Actually Want
Modern developers using agentic workflows want one, very simple thing: reliable access to models that can actually do the work required without introducing new issues into our codebase. In general, this means access to Opus 4.5 and 4.6, and to a lesser degree, Sonnet. No shade to the people working on Gemini 3, but it just isn’t there yet.
The last two years have shown that many of us are willing to pay a significant price for this service, as evidenced by your own Ultra tier and Anthropic’s USD $200 Max tier, but not all of us require that kind of heavyweight access. As you can see from the beginning of my post, I am hardly a power user. I’m not spinning up multiple autonomous agents simultaneously and I’m not using Opus, or even Sonnet, to implement every line of code. I still write a lot of code by hand. I’m not working on a codebase that spans thousands of files and millions of lines of code. My entire project is only about 3,000 lines of code, currently. Not only should I not be hitting my weekly rate limit after 2 days of moderate use, I shouldn’t have to pay power-user prices to access basic-user limits.
I would understand if the financial math worked out such that the Pro tier pricing is not enough to cover the kind of basic developer usage I’m talking about here. $30 is pretty cheap, considering how catastrophically expensive Opus is to run. But I also don’t care AT ALL about the other benefits of the Pro tier, like Whisk, Flow, NotebookLM, AI in my email, and Google Home. I will literally never use them, and I expect most developers will say the same.
We want an agentic AI tier, specifically for using code-capable models in Antigravity, that has a true 5-hour rolling reset instead of an arbitrary weekly cap.
For a tier like this, you could charge me double what I’m paying now and I would consider it fair. Hell, I would even be willing to pay 3x what I’m paying now, even if it made me a little salty. But I will never pay CAD $340/mo for services that I won’t get best use of, especially when you won’t tell us what “Highest rate limits to agent model in Google Antigravity” actually means. For all I know and can prove, the rate limits across all your plans are identical.
Last note on this topic: If you, or Anthropic, need to implement artificial weekly caps and misleading 5-hour resets because of a few bad actors abusing the system, then ban or limit those bad actors and stop punishing everyone else for their abuse.
Here’s What’s Going to Happen
I am going to get as much use out of the rest of my subscription month as possible, and if things haven’t been fixed by the end of it, I am going to cancel and go straight to Anthropic. I will abandon Antigravity completely and just use Claude Code on the $100 Max plan. Despite the price and ongoing issues that plan also has, it seems to be in a better state than the Google AI Pro plan, and it’s significantly cheaper than your Ultra plan.
My guess is that most Antigravity users will have a similar mindset. If we aren’t getting what we paid for and can’t do our jobs, why would we keep paying you? Why would we stay? You are going to experience a mass exodus of Antigravity users in the very near future.
And, not to put too fine a point on it, but you’re probably going to find yourselves on the wrong end of a class-action , too.