Google introduces “Thinking Budget” to control AI reasoning in Gemini 2.5 Flash

Google has launched a new AI reasoning control feature in its Gemini 2.5 Flash model, enabling developers to limit how much processing power the system uses when solving problems. Released on April 17, the so-called “thinking budget” aims to address growing concerns over inefficiency and high costs associated with advanced reasoning models.

The tool allows developers to set a precise limit on the model’s internal processing, ranging from zero to 24,576 tokens. This helps reduce unnecessary computation when handling simple queries—an issue that has driven up both financial and environmental costs in AI operations.

The shift reflects a growing industry challenge: while newer AI models deliver improved logical reasoning, they can overanalyze basic tasks, wasting resources. Google says this mechanism is designed to balance cost, performance, and sustainability.

AI researchers have noted similar inefficiencies across the sector, with models occasionally getting stuck in loops or using excessive computation without enhancing response quality.

This move also signals a shift from the trend of building ever-larger models toward a focus on efficiency and control. For developers and organizations, the ability to adjust reasoning depth offers a way to better manage both AI costs and environmental impact.