Google API Gemini: Enhanced Cost Control After Bug

Google recently announced the rollout of new tools aimed at better controlling the costs associated with using its Gemini API within the AI Studio platform. This announcement, made via a blog post on Monday, March 16, 2026, introduces several key features, including spending caps per project, a revamped usage tiers system, and new billing dashboards. These innovations come seven months after a billing incident that resulted in unexpected charges of several thousand dollars for some users.

Gemini API: Spending Caps and Enhanced Tiers System

One of the main new features introduced by Google is the Project Spend Caps system, which allows users to set a monthly spending cap per project directly within AI Studio. Project managers can activate, adjust, or disable this cap through the Spend tab in the interface. Once established, this cap remains active until it is manually modified or disabled. However, Google clarified that there is an application delay of about 10 minutes, during which any cost overruns remain the user's responsibility.

At the same time, Google has revamped its usage tiers (Usage Tiers) system, which determines the quotas and rate limits of the API. The changes include:

• A reduction in the spending thresholds required to access higher tiers,
• Automatic and immediate transitions between tiers as soon as the conditions are met,
• A monthly spending cap per billing account, adjusted based on the tier reached.

Additionally, Google has introduced several dashboards in AI Studio that offer:

• Detailed tracking of rate limits, including requests and tokens per minute,
• A daily breakdown of costs by project and model,
• Usage and error metrics,
• The ability to configure billing directly within the interface.

Response to a Major Billing Incident

The announcement of these new features was particularly anticipated due to an incident that occurred in August 2025. At that time, a bug in the billing system of the Gemini API generated abnormally high invoices for many developers. This issue was caused by a configuration error that billed internal tokens as image generation tokens, thus applying a high rate even for purely textual uses. Google acknowledged the problem and promised refunds, but the lack of capping mechanisms significantly exacerbated the situation.

One developer reported discovering charges of over $1,000 for image generation while he had only used the API to translate product descriptions, according to the site PPC Land.

In this regard, Google was lagging behind its competitors. OpenAI has long offered configurable spending caps per project, while Anthropic applies caps by tier and workspace. The Gemini API was until now the only major service in the industry not to provide such safeguards. These updates are part of a series of improvements made to Google's developer platform, following the recent launch of Gemini 3.1 Flash-Lite and Plan Mode in Gemini CLI.