Google has just deployed an improved version of its latest AI model, with a new feature allowing you to “turn reflection on or deactivate”.
Thursday, the technology giant deployed a first version of Gemini 2.5 Flash, an updated version of the 2.5 model which he published in March.
This model – a so -called “thought” model – has been nicknamed Google’s smartest to date, given its ability to reason through ideas before responding.
However, Google is now ready to let you choose how much this new model thinks. And if you really want, you can tell him to stop thinking completely.
In a blog article, the director of product management of Google Gemini, Tulsee Doshi, said that developers can “define budgets of reflection to find the right compromise between quality, cost and latency”.
The new feature aims to meet the intense processing and computer requirements of a new wave of “reasoning” models that have aroused interest in the AI industry, including O3 of Openai, published on Wednesday.
The new Google model aims to guarantee that its reasoning model only uses the treatment power that necessary and only applies it if necessary.
Doshi noted that not all tasks require the same reasoning. For example, the reasoning necessary to answer “How many provinces of Canada has?” is different from asking the AI to calculate the maximum bending stress on a cantilever beam of particular dimensions, she said.
To allocate different levels of reasoning capacity to user requests, Google will allow developers to define a “reflection budget” which, according to DOSHI, will offer “a fine grain control” on the number of tokens – data units – a model generates during operation.
The decision to introduce a “reflection budget” also follows a broader change in industry to become more “effective” in the use of calculation power.
This follows the publication of a model of reasoning in January of the Chinese startup Deepseek which claimed to use less computing power.
businessinsider
