Google has quietly overhauled how it tracks your Gemini AI usage, moving from simple daily request counts to a dynamic compute-based system. This shift, detailed in a recent support document, reflects the growing strain that advanced agentic features place on AI infrastructure. Whether you're a casual user or a power subscriber, these changes affect how much you can do—and how often. Here are the 10 most important things to understand about the new limits.
1. The Move from Request Counts to Compute-Based Usage
Gone are the days of counting individual prompts. Google now measures usage by the computational resources consumed by each interaction. Instead of a fixed number of queries per day, your limit depends on how much processing power your tasks require. This means a simple text request might cost little, while a complex session with image generation or extended reasoning could use up a larger share of your allowance. The shift mirrors broader industry trends as AI models become more resource-intensive.
2. What Determines Your Compute Consumption
Several factors influence how much compute you burn through per session. The complexity of your prompt plays a major role—asking Gemini to analyze a lengthy document or generate a detailed report eats more resources than a quick question. Features like image and video generation, Deep Research, and extended-thinking models (e.g., Pro or Deep Think) also increase usage. Additionally, longer chat histories that require the model to recall context add to the computational load. Google weighs all these variables to calculate your running total.
3. Tiered Limits Across Subscription Plans
Usage ceilings vary dramatically by plan. Free users get a baseline 'standard' limit, which is intentionally vague. Paid subscribers receive multipliers: Google AI Plus at $8/month offers double the standard limit, while AI Pro at $20/month provides quadruple. The top-tier AI Ultra plan at $250/month grants 20 times the standard allowance. This tiered structure rewards higher spenders with significantly more compute capacity, aligning costs with expected usage patterns for developers and businesses.
4. How the Rolling Refresh Works
Compute usage resets every five hours, but it's not a simple daily cycle. Instead, Google implements a rolling window that refreshes multiple times per day until you hit a weekly cap. For example, if you use half your allowance in a morning session, you'll get a partial reset five hours later, potentially allowing more use before the week ends. This design prevents users from exhausting limits in a single burst and encourages staggered usage over longer periods.
5. What Changed from the Old Daily Request Model
Previously, limits were straightforward: AI Pro users could send up to 100 Gemini Pro 3.1 prompts per day, regardless of complexity. A simple ask and a complex analysis counted equally. That system was easy to understand but inefficient for heavy users. The new compute-based model penalizes resource-heavy tasks while rewarding efficiency. For instance, a single multi-turn research session might now consume multiple 'requests' worth of compute, capping work differently than before.
6. Why Google Follows GitHub’s Lead on AI Credits
Google isn't alone in rethinking usage pricing. Less than a month ago, GitHub overhauled its Copilot plans, replacing 'premium request units' with 'AI Credits' tied to actual token consumption. Both companies are adapting to the reality that today's AI features—like autonomous agents that can spawn sub-tasks—consume vastly more resources than simple Q&A. The shift to compute-based limits is an industry response to the unpredictable, high-volume demands of modern generative AI workflows.
7. How Agentic Features Drive Up Compute Needs
The rise of agentic AI—models that can independently break down problems, execute sub-agents, and interact with tools—has fundamentally changed usage patterns. A single request to 'research and summarize the latest AI trends' might spawn dozens of internal operations, consuming tens of thousands of tokens across multiple turns. Fixed-rate plans weren't built for such bursts. Compute-based limits give providers a way to manage these resource spikes without raising base prices for everyone.
8. Anthropic Takes a Different Path (for Now)
Bucking the trend, Anthropic recently doubled Claude Code limits for its Claude Pro and Max plans. However, this move came only after securing a deal with SpaceX to boost its compute capacity. It's a reminder that infrastructure constraints—not just market strategy—drive these changes. Anthropic can afford generosity thanks to new hardware, while Google and others face scaling challenges with their massive user bases. The divergence highlights how compute availability shapes AI business models.
9. Anthropic Admits Plans Weren't Built for New Features
Anthropic’s executive conceded last month that existing Claude Pro and Max plans “weren’t built” for resource-hungry features like Claude Code and Cowork, which unleash AI agents on user PCs. This honesty underscores a universal challenge: AI plan pricing lags behind product capabilities. As models evolve, providers must retroactively adjust limits—often frustrating users who were sold on fixed promises. Compute-based systems attempt to future-proof against such mismatches.
10. Practical Implications for Users
For everyday users, the new system means more flexibility but also more uncertainty. You can't simply count prompts anymore; you need to be mindful of task complexity. Free users may hit limits faster if they experiment with advanced features. Paid subscribers, especially on higher tiers, gain headroom but still need to track usage. The rolling five-hour refresh helps moderate peaks, but planning your AI work in shorter, spaced sessions could maximize your weekly quota.
The shift to compute-based usage limits marks a pivotal moment in AI consumption pricing. Google’s move, aligned with industry trends, trades simplicity for fairness—heavy users pay proportionally more while light users retain affordable access. As agentic AI grows, expect more providers to adopt similar models. For now, understanding these new rules helps you get the most out of Gemini without surprises.