What makes Gemini’s 'Thinking' unique?

Gemini's 'Thinking' models are trained with Reinforcement Learning to utilize additional compute at inference time for more accurate answers^[1]. These models can spend tens of thousands of forward passes during a 'thinking' stage before responding to a query^[1]. This is integrated with other Gemini capabilities, such as multimodal inputs and long context, where the model decides how long to think before answering^[1].

Users can also set a Thinking budget, which constrains the model to respond within a desired number of tokens, allowing for a tradeoff between performance and cost^[1]. The Gemini 2.5 Thinking models are the most well-rounded reasoning models to date^[1].