Google’s latest update to Gemini takes on OpenAI’s reasoning AI model

Google’s Gemini AI has been steadily evolving throughout the year, with updates that signal its ambition to compete in the crowded AI market. Earlier this month, the tech giant introduced Gemini 2.0 Flash, a model which it says is designed for faster and more efficient multimodal generative tasks.

Now, Google is taking the Gemini 2.0 Flash capabilities a step further with an experimental reasoning model, Gemini 2.0 Flash Thinking, which aims to tackle complex problems in areas like math, physics, and programming.

This multimodal reasoning model, now available in Google’s AI Studio, is positioned as a “first step” in what the company calls its reasoning journey. Logan Kilpatrick, AI Studio’s product lead, described the model as an early experiment in using “thoughts to strengthen reasoning.”

According to him, like most AI reasoning models, Gemini 2.0 Flash Thinking attempts to fact-check itself, breaking problems into smaller tasks before arriving at a solution. For instance, the model can provide step-by-step explanations for solving complex equations—somewhat mimicking human-like problem-solving—though its accuracy can still falter, as early tests have shown it struggling with simpler tasks like counting letters.

This update also signals Google’s growing competition with OpenAI. The ChatGPT creator released the full version of its 01 reasoning models just a few weeks ago and recently introduced a successor–the o3 Model–which it claims is more powerful at AI reasoning capabilities.

However, both Google's and OpenAI's reasoning models share a significant challenge: they require substantial computational power which leads to slower response times. For Google to truly stand out in this space, it would need to deliver a reasoning model that surpasses OpenAI's o1 capabilities at a significantly lower cost than the $200/month price tag OpenAI charges for its top-tier reasoning model.

But since Gemini 2.0 Flash Thinking is still in its experimental stage, it’s too early to tell what the cost would be. And seeing as reasoning models often require higher operational costs, it'll be interesting to see how Google prices its model.

For now, though, Google says it is heavily invested in this space with over 200 researchers dedicated to advancing its reasoning AI. With companies like Alibaba and DeepSeek also racing to refine their reasoning AI, the stakes are high for Google to show that its multimodal capabilities and structured approach to problem-solving can lead to practical, scalable applications.