Google has overhauled how its Gemini AI platform bills for usage this summer, introducing a more granular pricing structure that varies by model, task, and data volume, changes that mean users may get fewer responses for the same spending level unless they actively track their consumption.
The updated pricing divides costs into several components. Input tokens represent the text, images, or files sent to the model; output tokens represent the model’s response. Each Gemini model carries a different rate, and optional features such as larger context windows add further cost. Grounding with Google Search, for example, adds a per-query surcharge of $35 per 1,000 prompts for Gemini 2.5 and 2.0 models beyond free daily limits, and $14 per 1,000 for Gemini 3 models.
Google has also restructured its consumer subscription tiers. The new entry-level Google AI Plus tier, introduced in July 2026, offers 400 gigabytes of storage and double the free-tier usage limits for $7.99 per month. Above it, Google AI Pro costs $19.99 per month (formerly Google One AI Premium, now with 5 terabytes of storage) and Google AI Ultra starts at $99.99 per month with 20 terabytes of storage and access to Deep Think capabilities. A $200-per-month premium tier within Ultra adds additional capacity.
For developers, the Gemini API remains pay-as-you-go. Gemini 3.5 Flash costs $1.50 per million input tokens and $9 per million output tokens. Gemini 2.5 Pro costs $1.25 and $10 respectively. An agent-pricing SKU called AlphaEvolve, added in July 2026, charges a transparent 2x surcharge over the base model rate, making the all-in cost exactly three times the underlying model price.
The billing impact is easy to underestimate. Ten thousand requests with 2,000 input tokens and 1,000 output tokens each consume 30 million tokens. Even at low per-token rates, the total adds up quickly. Using a premium model unnecessarily, what one analyst described as “taking a taxi instead of a bus”, amplifies the effect.
Google provides several tools for tracking usage. The billing console shows spending by project and model. Usage pages break down input and output token counts. API logs offer per-request granularity. Setting budget alerts at 50, 80, and 100 percent of monthly caps is recommended, as is tagging projects by use case, chatbot support, coding assistance, document search, to identify which feature drives the highest consumption.
Small changes in prompt structure can yield significant savings. Cutting each prompt by 25 percent in a system processing 25,000 daily conversations with an average of 800 input and 500 output tokens saves millions of tokens per month without reducing chat volume. Matching model strength to task complexity, using cheaper models for classification and tagging while reserving premium models for complex reasoning, is another lever.
“Google’s AI pricing is manageable if you measure it closely, choose the right model, and set hard spending alerts,” the company advises.
Sources: How Google’s New Gemini Rates Work and How to Track Your Usage (Wired, July 2026); Google Gemini Pricing (UsagePricing, July 2026)