Google’s Shift to Compute‑Based Gemini Limits: What Changed and Why Users Pushed Back

This means a short text query may consume very little quota, while a long conversation using advanced tools or coding assistance can drain it quickly. Google says this approach better reflects the reality that some AI tasks require far more computing resources than others.

How the New Limits Affect Free and Paid Plans

The compute‑based system applies across Gemini’s subscription tiers, but each plan receives a different usage budget.

According to Google’s support documentation:

• Free users: receive standard compute limits.
• AI Plus subscribers: get roughly 2× higher limits than the free tier.
• AI Pro subscribers: receive about 4× higher limits than the free tier.
• AI Ultra subscribers: can receive dramatically higher allowances depending on the product and features used.

At Google I/O 2026, the company also introduced the $100/month AI Ultra plan, which includes significantly larger usage limits—such as a five‑times higher limit in the Antigravity coding environment compared with AI Pro.

In other words, the key differentiator between plans is no longer just feature access—it is the amount of compute budget available to the user.

Why Users Hit Limits So Quickly

The immediate problem was predictability. Under the old system, users could roughly estimate how many prompts they had left. With compute‑based quotas, that predictability largely disappeared.

Because usage depends on task complexity, heavy workloads can burn through quota much faster than expected. Examples include:

• Long coding sessions
• Agent‑style workflows
• Media generation or advanced reasoning features

Reports from developers and users showed that intensive sessions could exhaust the five‑hour allowance or even the weekly cap after only a few work sessions.

Many subscribers felt blindsided by the change. Some described the new limits as a "bait‑and‑switch," arguing that their paid plans now delivered less practical usage than before—even though Google had technically not reduced the official subscription tiers.

Google’s Emergency Fix: Raising Limits

The backlash prompted a rapid response from Google, especially within Antigravity, the company’s AI‑powered coding tool.

Within days of the rollout:

• Google tripled Gemini model rate limits in Antigravity.
• It reset weekly quotas for users so they could continue working immediately.

Shortly afterward, the company tripled limits again, further expanding the available compute budget for developers using the tool.

The quick adjustments suggest Google underestimated how quickly real‑world workflows—particularly coding and agent‑based tasks—would consume the new quotas.

What This Episode Reveals About the AI Economy

The controversy highlights a growing tension across the AI industry.

Modern AI systems do not have uniform costs. A short text reply might be cheap to generate, while long‑context reasoning, coding agents, or video generation can require dramatically more compute resources. That makes simple “message count” limits increasingly unrealistic for providers to sustain.

Compute‑based quotas solve the economic side of that problem—but they introduce a usability challenge. When limits depend on opaque calculations about complexity, users may struggle to predict how much access they actually have.

Google’s Gemini rollout illustrates this trade‑off clearly: a system designed to align usage with compute cost ended up confusing users and triggering backlash almost immediately.

For AI companies scaling large models, the challenge going forward is balancing three competing forces:

• Rising inference costs for advanced AI features
• Clear, predictable limits that users understand
• Subscription plans that still feel valuable to paying customers

As AI assistants become more powerful—and more compute‑intensive—that balancing act is likely to become one of the defining product challenges of the industry.