A brief look at Google's strategic shift following the Google AI Pro quota reduction
I originally intended to write a post to vent a day or two after the Antigravity quota reduction occurred, but I was busy with other things over the weekend. By the time I wanted to start writing this week, I felt that so much time had passed that it might look like I was just jumping on the bandwagon. However, after experiencing issues where I couldn't use the Claude model despite having remaining quota, and seeing more news trickling in, I decided it was worth putting my thoughts together.
The Incident
On March 12, many users opened Antigravity in the morning to find their Gemini Pro/Claude model quotas reset to 0, despite having quota left the day before. Looking at the reset time, it appeared to have triggered a rolling weekly quota limit. At the time, the community was debating whether this was a bug or if Google was further tightening quotas.
To be fair, everyone somewhat expected the quotas to tighten. After all, the previous model—with a 5-hour cooldown, separate quotas for each family account, and different limits for various services—was a burn-rate strategy that couldn't last forever. Everyone was just using it while they could. Coupled with the "buy a phone, get a half-year subscription" promotion in the second half of last year, and the half-price annual Pro subscription offer at the end of the year, the user base grew significantly, making subsequent tightening inevitable. Therefore, when the rolling weekly quota for the Claude model was introduced in late January, most people took it in stride (or so I assume).
However, this latest round of limitations was somewhat unexpected. Other providers at the $20 tier do not restrict high-end model usage to this extent.
In fact, signs of deteriorating experience appeared even before the March cuts. Starting in February, some users reported that Gemini felt "dumber," suspecting that the models were being quantized or that resources were being throttled. At the time, I replied that I didn't notice much difference, but by mid-February, I started to feel that something was off. Antigravity began struggling with tasks that took a long time but failed to complete properly. At first, I suspected I hadn't explained things clearly, or that some steps were inherently complex and beyond its capabilities.
It wasn't just the model quality that was declining; the rate of quota consumption also accelerated significantly. Sometimes, the Claude model would burn through the 5-hour quota in just 2–3 tasks. After burning through two such cycles, the weekly quota would be down to 60%. In other words, users were hit by a double whammy: "weaker models" and "less quota."
Following the reduction, Google issued a statement clarifying the positioning of each plan:
We're evolving Google AI plans to give you more control over how you build. Every subscription includes built-in AI credits, which can now be used for Antigravity, giving you a seamless path to scale.
Google AI Pro is the home for the practical builder, hobbyists, students, and developers who live in the IDE and don't necessarily rely on an agent. This plan features generous limits for Gemini Flash, with a baseline quota included to "taste test" our most advanced premium models.
Google AI Ultra serves as the daily driver for those shipping at the highest scale who need consistent, high-volume access to our most complex models.
If you're on Pro but need "extra juice" for a heavy sprint or deeper access to premium models, simply top up your AI credits to customize your plan.
Keep building. Keep shipping.
In short: The Pro plan is positioned around Gemini Flash as the primary workhorse, with advanced models (Gemini Pro, Claude) intended only for "taste testing." Ultra is the plan for those who need to use advanced models heavily. If Pro users need more quota temporarily, they can purchase additional AI Credits.
About AI Credits
I don't have an issue with using AI Credits to top up model quotas. After all, the 1,000 credits provided monthly were originally intended for image and video generation in Google Labs, but without a clear concept or need, I didn't know what to generate. Being able to use them to supplement model quotas is actually more practical.
It is worth noting that on the same day as the quota reduction, Google published How we're reimagining Maps with Gemini, announcing the integration of Gemini into Google Maps with conversational search ("Ask Maps") and a major navigation visual upgrade called "Immersive Navigation," initially launching in the US and India. Compressing quotas for developer tools while pouring AI capabilities into consumer products—when these two events are placed side-by-side, the strategic direction becomes clear.
On March 18, Ryan J. Salva, Senior Director of Product for Google Code Assist, explained in Service update: mitigating abuse and prioritizing traffic #22970 that free users of the Gemini CLI are restricted to Gemini Flash, while Pro-level access or higher is required for Gemini Pro. Furthermore, traffic priority is determined by account status and license type, with paid users prioritized and free users throttled. I suspect this was primarily aimed at those who had previously registered large numbers of accounts for use with OpenClaw.
On that same day, my Antigravity Claude model was completely unusable despite having a full quota, stuck in a loop of "Working" and "Generating" messages. It seems it wasn't just a model quota issue; the server resources allocated to the models might have been reduced as well.
On March 19, Google AI Studio announced the "Build" feature. I found a more comprehensive application in this article: AI Studio officially supports full-stack development! Google launches Vibe Coding to let you complete frontend and Firebase builds with voice commands. It allows users to create a website including a Cloud Firestore database and OAuth functionality using commands. However, one should verify the billing status of Firestore on GCP to avoid unexpected charges.
Taken together, these moves point to a single trend: Google is reallocating AI resources from standalone developer tools (Antigravity, Gemini CLI) to integrated applications within its own ecosystem (Maps, AI Studio Build). For Google, making AI a built-in capability of its own products has more strategic value than maintaining a high-quota third-party developer tool. The strategic focus is shifting from consumer-facing developer users to a "use-as-you-go" developer system centered on ecosystem integration.
Is it still worth subscribing to Google AI Pro?
| Feature | Plus | Pro | Ultra |
|---|---|---|---|
| AI Credits (Shared Pool) | 200 / month | 1,000 / month | 25,000 / month |
| AI Assistant | |||
| Gemini 3.1 Pro | ✓ | ✓ | ✓ |
| Deep Research | ✓ | ✓ | ✓ |
| Deep Think Reasoning Model | — | — | ✓ |
| Gemini Agent | — | — | ✓ (US only, English only) |
| Google Search AI Mode (Gemini 3 Pro) | ✓ | ✓ | ✓ |
| Search Labs Experimental Features | ✓ | ✓ | ✓ |
| AI Phone Calls | — | ✓ (US only) | ✓ (US only) |
| Use Gemini directly in Google Apps | ✓ | ✓ | ✓ |
| Creative Generation | |||
| Video Generation | Veo 3.1 Fast | Veo 3.1 Fast | Veo 3 |
| Photo to Video | ✓ | ✓ | ✓ |
| Google Photos Magic Editor | — | ✓ (US only) | ✓ (US only) |
| Image Generation (Nano Banana Pro) | ✓ | ✓ | ✓ |
| Flow Video Creation | ✓ | ✓ | ✓ |
| Lyria 3 Music Generation | ✓ | ✓ | ✓ |
| Learning Tools | |||
| Multi-page Reports | ✓ | ✓ | ✓ |
| NotebookLM | ✓ | ✓ | Strongest Model |
| Programming | |||
| Jules Asynchronous Code Agent | — | ✓ | ✓ |
| Gemini CLI / Code Assist IDE | — | ✓ | ✓ |
| Google Antigravity Agent Model | — | ✓ | ✓ |
| GCP Credits (Google Developer Program) | — | US$10 / month | US$100 / month |
| Additional Services | |||
| Cloud Storage | 200 GB | 2 TB | 30 TB |
| Family Group Sharing | Up to 5 people | Up to 5 people | Up to 5 people |
| Google Home Premium | — | ✓ | — |
| YouTube Premium | — | — | ✓ |
Comparing the table, you can see that the main difference between Plus and Pro lies in the programming features. As for Ultra, I'll skip it; the Gemini model's strength in programming is primarily UI design, and it doesn't stand out in other aspects compared to Claude or ChatGPT. It's unlikely anyone would subscribe to this tier just for coding, especially when it costs an extra $50 (the top tiers of the other two providers are $200, with Claude Max having an entry tier at $100).
So the question returns to: Is the extra programming functionality in Pro worth the price difference over Plus? If you only use an agent to assist with development and don't use it to work on side projects after hours, the $20 plans from most providers are generally sufficient. But if you want to use Vibe Coding as your primary development method, the quota requirements are completely different. The $20 plans from any provider aren't enough for continuous use. Antigravity used to manage by switching between multiple accounts, but that was before the January quota limits.
After the March 12 adjustment, it's a different story. On one hand, the Gemini Pro and Claude model quotas for Google Pro are already lower than equivalent plans from other providers; on the other hand, the aforementioned issues with declining model quality and service instability haven't improved. With low quotas, poor quality, and insufficient stability, it's really better to switch to another provider at the same price point. Over the past two weeks, I've heard more and more people switching to Claude Code or ChatGPT Codex.
Unless Google makes significant adjustments to its plans later, I wouldn't recommend subscribing to Pro under the current circumstances. For non-programming ecosystem users, Plus and Pro are identical in usage; for programming, Pro is currently too lackluster. If you need it, you're better off subscribing elsewhere. In fact, over the last two weeks, opening the Gemini App keeps triggering prompts to upgrade to Ultra, which makes Pro feel like an awkward middle ground even for Google itself.
Change Log
- 2026-03-20 Initial document creation.