Google Ai Pro Subscription Quota

Google Antigravity
, ,
1

I have used Antigravity for a few years. Over the past weeks my quota drains within a few prompts. My usage has not changed.

Three problems stand out.

  1. Repeated file reads. The agent opens the same file five or six times in one task. Each read costs quota and adds no new information.
  2. Looping. The agent repeats steps without progress. A single task burns through quota before any output arrives.
  3. Slow responses. Latency has grown a lot. Tasks that finished quickly in the past now stall for long stretches.

I also pay for Claude Pro. I run many projects and reach its limit often, but each prompt returns finished work, so the spend feels fair. I would pay for the Max tier without hesitation. Antigravity spends quota on repeated reads and loops, so the limit arrives with little to show.

Google built the hardware, the research, and the early tooling for this field. Users expect a better result from you. Please fix the loop behavior, cache file reads within a session, and show quota usage per action so users see where quota goes.

I want to keep using Antigravity. Right now the cost per task makes this hard…

im working on a project right now and i wanted to make like a simple UI change and its gone through half my quota…

2

I have had a similar issue. End of last week I got the famous self-shame regret loop, burned through a good share of my quota. Heard nothing at all about that bug report, but since then I’ve been having some wild issues.

I’ve reported it multiple times. A process I used teamwork_preview before on took TWO AND A HALF 5 hour windows to complete. It was a part of a project divided into stages. Before these issues it took 30-40% of my 5 hour window. Now this code part was a bit bigger, I was okay with 50 to 60%. but 250?

Then a few hours later I asked two simple queries on 3.8 flash medium Boot up my local postgreSQL - the agent has a local environment file which it immediately found, with instructions on how to do this - this is already slow, it feels like it’s doing extra work it shouldn’t. Then ask it to spin up my local R3 spoof. Okay it needs to search more files, do a grep, sure.

2 queries on 3.8 medium. Before this they would be done in <10 seconds. Now, 5-10 minutes. Whatever. 5 hour window… 35% gone?!

I’ve reported the convos, I have no idea whats going on. Even /btw’s take me 5% regularly now on 3.8 low/mid. and to be clear my customizations window is 6% filled for this project, and all these were in fresh conversations in the project.

3

Hi @ADM1SH,

To help us investigate and resolve this issue accurately, could you please provide:

  1. Diagnostic Logs (`.txt` file) or Error Screenshot: If you encountered an error, please attach either your diagnostic logs dumped in a `.txt` file (from `Help > Toggle Developer Tools > Console` or `View > Output > Antigravity Language Server`) or an error screenshot.
  2. Detailed Explanation of the Issue: What specific problem or unexpected behavior did you encounter, what were you trying to do, and what happened vs. what you expected?
4

Hi, thanks for the response. Here are the details you asked for.

Diagnostic Logs

I’m attaching antigravity_language_server_diagnostic_logs.txt, pulled from View > Output > Antigravity Language Server. Environment:

Application: Google Antigravity IDE
IDE Version: 2.15.1
Language Server Version: 1.2.8
Language Server PID: 20939
OS: macOS Darwin arm64 (Apple M4 Pro)
Active Model: Gemini 3.8 Flash (High)
Cloud Code Endpoint: https://daily-cloudcode-pa.googleapis.com
Log Source: ~/.gemini/antigravity-cli/log/cli-20260923_091215.log

Relevant excerpt from the language server output:

I0923 09:12:15.276925 82 server.go:1573] Starting language server process with pid 19057
I0923 09:12:15.277713 82 server.go:1624] Language server version: 1.2.8
I0923 09:12:15.278573 82 server.go:624] Language server listening on random port at 50065 for HTTPS (gRPC)
W0923 09:12:15.666029 103 cache.go:135] Cache(loadCodeAssistResponse): Singleflight refresh failed
E0923 09:12:15.666726 103 errorreport.go:224] Failed to poll ListExperiments
I0924 19:17:41.090243 807244 http_helpers.go:305] URL: https://daily-cloudcode-pa.googleapis.com/v1internal:streamGenerateContent?alt=sse
I0924 19:17:41.249334 417 quota_manager.go:45] doRefreshQuota: starting reload (force=false)
I0924 19:24:54.735935 417 quota_manager.go:45] doRefreshQuota: starting reload (force=false)

Detailed Explanation

I’ve used Antigravity for a few years. Over the past few weeks my quota drains within a few prompts, and my usage habits haven’t changed.

What I was trying to do: a minor UI change to a small component file in an active codebase.

What I expected: the agent reads the target file once, applies a targeted diff, and returns the finished edit quickly, with quota use proportional to a small change.

What actually happened: the agent read the same file five or six times in one task with no new information gained from the repeat reads. It then cycled through repeated steps without moving the task forward. Response latency stretched out for long periods where past sessions would have finished fast. By the time I got a working answer, the task had used up about half my total quota, on a UI tweak that should have cost very little.

Three problems stand out for me:

Repeated file reads. The agent reopens the same file multiple times in one task. Each read spends quota and adds nothing new.

Looping. The agent repeats steps without progress. A single task can burn through quota before any output arrives.

Slow responses. Latency has grown a lot recently. Tasks that used to finish quickly now stall for long stretches.

For comparison, I also pay for Claude Pro and run several projects through it. I hit its usage limits often too, but every prompt comes back with finished work, so the spend feels fair. I’d move to a higher Antigravity tier without hesitation if the quota reflected real output. Right now quota goes into repeated reads and loops instead, so the limit shows up with little to show for it.

Google built the hardware, the research, and a lot of the early tooling in this space. I’d like to see Antigravity match that. My ask: fix the loop behavior, cache file reads within a session, and show quota usage per action so I can see where it’s going.

I want to keep using Antigravity. Right now the cost per task is what’s making that hard.

5

image
image802×429 13 KB

Same problem. I used the 3.7 flash model, it read the same file more than 10 times, and in the end it couldn’t do what was requested — it’s complete nonsense.