Google Enters Reasoning AI Race With Experimental Gemini Model
Google introduces Gemini 2.0 Flash Thinking Experimental, a new AI model designed to work through complex problems by fact-checking itself. However, early testing shows the experimental system still struggles with basic tasks like counting letters.
Google releases a new experimental reasoning AI model called Gemini 2.0 Flash Thinking Experimental through its AI Studio platform. The model trains to use internal thoughts to strengthen its reasoning over complex problems in programming, math, and physics. Google executives describe this release as the first step in the company's reasoning journey, noting that increasing the computing time during inference yields promising results.
Similar to OpenAI's o1 model, this system effectively fact-checks itself to avoid common AI pitfalls, although this extra verification means it takes longer to generate answers. When given a prompt, the model pauses to consider related questions and explains its reasoning before summarizing the most accurate solution. This self-reflective approach aims to tackle highly complex, multimodal challenges that standard models easily miss.
Despite its advanced design, early testing reveals significant room for improvement in this experimental phase. For example, the reasoning model still fails at basic tasks like accurately counting the number of specific letters in a simple word. The launch comes as rival AI labs also rush to release their own reasoning models to compete in this rapidly evolving segment of the artificial intelligence industry.