On Thursday, Google unveiled a new artificial intelligence (AI) model within its Gemini 2.0 family, specifically designed for enhanced reasoning capabilities. Named Gemini 2.0 Thinking, this advanced large language model (LLM) allows for extended inference times, enabling the system to dedicate more time to problem-solving. According to the Mountain View tech company, the model excels at tackling complex reasoning, mathematics, and coding tasks while also operating at a greater speed despite the increased processing duration.
Google Introduces New AI Model with a Focus on Reasoning
Jeff Dean, Chief Scientist at Google DeepMind, announced the new Gemini 2.0 Flash Thinking AI model in a post on X (formerly known as Twitter). He emphasized that this LLM is “trained to use thoughts to strengthen its reasoning.” The model is now accessible through Google AI Studio, and developers can utilize it via the Gemini API.
Gemini 2.0 Flash Thinking AI model
Members of the Gadgets 360 staff had the opportunity to test the AI model and observed that the Gemini model, focused on advanced reasoning, effectively handles complex questions that challenge the capabilities of the 1.5 Flash model. During testing, typical processing times ranged from three to seven seconds, marking a notable improvement over OpenAI’s o1 series, which can take more than 10 seconds per query.
The Gemini 2.0 Flash Thinking model also provides insights into its reasoning process, allowing users to view how the AI arrived at its conclusions and the steps involved. It successfully identified the correct solution in eight out of ten attempts. Given its status as an experimental model, some errors during operation are to be expected.
Although Google refrained from disclosing specific architectural details about the AI model, it outlined its constraints in a developer-oriented blog post. Currently, the Gemini 2.0 Flash Thinking has an input limit of 32,000 tokens and accepts only text and images as inputs. The output is restricted to text and capped at 8,000 tokens. Furthermore, the API does not include built-in tool functionalities such as search or code execution.
