I have Claude max 5x, and it feels like the quota is just a little bit less than ultra’s Claude. So 20x should be much more than antigravity, I guess.
2 Likes
That’s how its supposed to work by now. However, in my particular workflow, it consumes ~2.5k credits an hour . I’m mostly analizing and refactoring microcontrollers code. ( RTOS, APIs , SDKs , low level Linux,OSX, Windoze ).
Quick math → I eat my quota in an hour or so , 3.5 is then ~9k credits . I use all the 25k credits in a day or two of being easy with my keyboard.
Before the “update” I was able to work on two different projects at once without hitting any limit.
So it’s clearly a deprecation of the service without any prior notice, Ultra is not cheap tho , and it is not really deterministic.
2 Likes
I would love to hear more from your tests. Planning on switching to CC soon.
How’s Claude Code’s Max 5x? No way I’m gonna keep paying for this, running 2 ultra accounts and getting locked out in 1 hour of usage total. Also, just got hit by a 3d limit in one of the accounts, love that cherry at the top of my cake :). Disputing for a refund .With that money I’ll just do Max 20x and whatever
3 Likes
Your response seems to be targeted towards users who are currently on the Google AI Pro plan. The issues in this thread that I and others are having are on the Google AI Ultra plan.
The amount I can actually use in a 5 hour period has dropped by about 80 to 90%.
Is there any official update on the quota changes for the Ultra users?
6 Likes
I’d like some real clarification on what “generous usage limits” actually means from a technical standpoint.
You position Antigravity as a tool for practical builders, so we expect concrete numbers: actual quotas, token limits, and exact pricing.
For example, if a request uses 2,000 tokens, what’s the precise quota impact or cost? Without transparent metrics, it feels more like a guessing game than a reliable engineering product.
This lack of clarity makes it hard for us to properly evaluate Antigravity for real-world use, it just adds friction, delays, and uncertainty to our workflows.
Update: I’ve switched over to Codex, the workflow feels much smoother and more seamless than with AG.
2 Likes
4 Likes
I guess the only way to know is to try it … probably do that after my projects
1 Like
max 5x should be lesser because the cost is lesser than AG ultra. we need to have a better unit/cost measure.
Yeah, because of Google’s black box quota limits, this is purely my subjective comparison between Antigravity and Claude code. Honestly, no developer needs a development tool with completely opaque quotas.
Hi Antigravity Team,
I am writing to report a critical and frustrating inconsistency regarding quota consumption for Claude models under the Google AI Ultra subscription, specifically affecting users in the UTC+8 time zone. This issue appears to have started earlier this week.
The Core Issue:
Since the beginning of this week, I have observed that initiating just a single conversation session with the Claude model (either Sonnet or Opus) within Antigravity immediately consumes approximately 60% of my total quota.
This completely defeats the purpose of the Ultra subscription. I upgraded to the Ultra tier specifically to access significantly larger capacity for heavy development workflows, not to have my entire allowance exhausted by a single chat session. If one conversation wipes out more than half of my quota, the “Ultra” label is misleading, and the service becomes practically unusable for any real-world tasks.
Comparative Observations:
-
Ultra vs. Pro Paradox: Paradoxically, the quota behavior for Claude on the Pro subscription seems far more reasonable. It is absurd that an Ultra subscriber effectively receives less usable utility per dollar than a Pro subscriber. Currently, the Ultra subscription’s Claude quota is performing worse than the Pro tier.
-
Gemini Stability: In stark contrast, the Gemini models (including Gemini 3 Pro) on the same Ultra account are functioning correctly. Their quota consumption aligns with expectations, allowing for extended coding sessions without premature exhaustion. This confirms the issue is isolated to the Claude integration on the Ultra plan.
Discrepancy with Value Proposition:
The current behavior contradicts the fundamental promise of the Ultra tier: high volume usage. Paying a premium for “Ultra” implies the ability to run multiple long-context sessions or complex agents. Having 60% of the quota vanish after one prompt suggests a severe bug in calculation logic or a misconfiguration specific to UTC+8 regions.
Request:
Could the team please provide an immediate explanation?
-
Is this a known bug causing massive over-charging for UTC+8 users?
-
Why does a single chat consume 60% of the quota when the expectation was to support dozens of such sessions?
-
Why does the Pro tier currently offer better effective capacity for Claude than the Ultra tier?
I chose the Ultra plan specifically for its promised scale. The current situation, where a single interaction drains the majority of my resources, is unacceptable. I hope for a swift investigation and a fix to restore the service levels I paid for.
Environment Details:
-
Subscription: Google AI Ultra
-
Region/Timezone: UTC+8
-
Affected Models: Claude (Sonnet/Opus)
-
Working Models: Gemini Series
-
Specific Symptom: ~60% quota consumed per single conversation start.
-
Onset Date: Earlier this week (approx. March 22–25, 2026)
Thank you for your urgent attention to this matter.
Best regards
8 Likes
Very well said, the reduction went even crazier today, it’s a fraction of a previous day every single day for the past 3 days! This is totally unacceptable and a business morality disgrace.
1 Like
so … the issue might be on Claude’s side?
lesson is that we should not rely on a single provider. There are other newer models that are catching up and costs much less like Minimax 2.7 and MiMo V2 Pro.
3 Likes
just restrict quota with no announcement is crazy. paid for the service and it suddenly goes crazy. feel like i just got scammed.
4 Likes
274.99 freaking Euros per month for Google changing quotas without any kind of warning!
This is inadmissible, and I’m pretty sure this is even illegal in European Union!
16 Likes
Ultra is not worth the money. Cancelled my supscription
6 Likes
yeah if they dont give explaination so do i.
cant trust on unstable service like this.
3 Likes
The quota change for ultra is shocking…. 1 hour of use on a single project eats it completely. I was comfortably working on three different projects before this and never hit the cap. I just downgraded back to pro after 3 months of ultra - was good while it lasted!
1 Like
I’m an Ultra user too, and not even a heavy user. I only ever run one project at a time with Claude Opus 4.6. My quota used to stay at 100%, or at worst 80%, but these past few days it’s been dropping to 60% in just over an hour.
I’m basically paying for the Ultra plan only because of Antigravity. If the cost is close to Claude Code but the quota is much lower, then what’s the point? At that stage, I might as well just switch to Claude Code, especially since Antigravity doesn’t really offer enough extra value over Claude
2 Likes
Why not expand the model roster to include LLMs like MiniMax 2.7 or MiMo v2 Pro? AG already set a great precedent by adding Claude and GPT-OSS 120B. Giving users more model options prevents them from getting stuck with underperforming models for their specific use cases. It’s a win-win: AG retains its user base and can focus purely on improving the harness, rather than users feeling forced to look elsewhere for the models they need.