The Gemini digital environment seems to be taking advantage of the last days of the year to catch up. After the presentation of the Gemini 2.0 model, it surprises us with the launch of Gemini 2.0 Flash Thinking, its first model focused on “reasoning” in AI.
This new experimental model promises to transform how machines approach complex problems, offering better understanding, analysis and decision-making compared to traditional AI models.
But what makes this model unique and how does it differ from other proposals on the market?
What is Gemini 2.0 Flash Thinking?
Gemini 2.0 Flash Thinking is an experimental version of Google’s Gemini 2.0 AI model, specifically designed to improve logical reasoning and the ability to decompose complex problems into simpler logical steps.
Unlike conventional AI models that directly answer questions without showing their internal process, this model has the ability to “think out loud”, that is, it breaks down its reasoning while solving problems.
This approach not only allows for a better understanding of the final answer, but also provides transparency to the process, which is crucial to avoid errors or “hallucinations” that often occur in AIs.
Not yet available as a public chatbot
The model is available through Google AI Studio, Google’s AI prototyping platform, and Gemini API, where developers can interact with it to generate more informed and logical responses.
Additionally, it offers outstanding performance in areas such as programming, mathematics, physics, and other complex problems that require in-depth analysis.
How does Gemini 2.0 Flash Thinking work?
The operation of the model is based on the ability to reason, which differentiates it from other models that simply generate answers.
When posed a question, Gemini 2.0 Flash Thinking does not respond immediately. Instead, you go through a pre-thinking process in which you evaluate several possible answers, consider the contexts, and do a step-by-step breakdown of how you would arrive at a logical conclusion.
During this process, the model explains its reasoning, allowing the user to understand how the final answer was arrived at.
This model is optimized to work with multimodal input, meaning it can process both text and images, a feature that further strengthens its ability to understand complex problems.
However, due to its reasoning approach, the model tends to be slower than other AI models in terms of response time, as it must go through these thought processes before providing a response.
Chain of reasoning or Chain of thought (CoT)
One of the most interesting aspects of Gemini 2.0 Flash Thinking is its ability to display your thoughts as you solve problems. Additionally, you can break down a larger problem into smaller, manageable logical steps.
These intermediate steps are known as a chain of reasoning, or chain of thought in English. This can improve the ability of language models to reason in a complex way, similar to how human beings do.
This is especially useful in tasks that require detailed analysis, such as solving mathematical problems, explaining scientific theories, or making decisions based on multiple factors.
Differences with the OpenAI o1 model
One of the direct competitors to Gemini 2.0 Flash Thinking is OpenAI’s o1 model, which is also designed to perform advanced reasoning. Although both models share a similar approach to reasoning and complex problem solving, there are key differences in how they work.
In the absence of tests and benchmarks, the main difference is in accessibility and customization.
Gemini 2.0 Flash Thinking offers more granular control over security settings through Google AI Studio, allowing users to adjust things like moderation of dangerous or explicit content.
This type of customization is not as easily accessible in o1, which may be a plus for those who want more control over the responses generated by the AI.
Gemini 2.0 Flash Thinking is accessible in chatbot form, to the general public through Google AI Studio for free (at the time of writing). Unlike OpenAI o1, which is only accessible with a paid ChatGPT subscription (Plus, Teams, etc). Although o3 has been presented, it is still not available for subscriptions in December 2024.
Competition in reasoning models
Throughout 2024, we have witnessed a race to develop AI models that can reason in advanced ways.
In addition to Gemini 2.0 Flash Thinking and OpenAI’s o1, several competitors have introduced similar models seeking to improve AI reasoning and analysis capabilities.
DeepSeek-R1
One of the most recent competitors in this field is the DeepSeek-R1 model, launched by DeepSeek, an AI research company backed by quantitative traders.
This model is designed to tackle complex tasks and act as a “reasoning agent,” similar to models from OpenAI and Google, and is one of the leading contenders in the field of advanced reasoning.
Qwen from Alibaba
Another important model is Qwen QwQ-32B, launched by Alibaba’s AI team in November 2024.
Qwen was presented as a direct competitor to OpenAI’s o1, standing out for its reasoning capacity and its focus on multimodal artificial intelligence, which places it in the same category as Google’s model.
An interesting 2025 awaits us
The release of Gemini 2.0 Flash Thinking marks another milestone in the emerging era, one in which machines not only solve problems, but also explain how they do it.
By offering a reasoning model, Google seeks to dispute the leadership currently claimed by OpenAI with its o1. However, both models are still experiments, since they only offer comparative advantages in specific problems, being much slower in the majority.
With more competitors entering the game, we are confident that 2025 will be a key year for the development of advanced reasoning models that will change the way we understand and use AI. And how AI understands us.
This post is also available in: