Two weeks ago, Google and Openai postponed their model award -winning performance in the International Mathematics Olympiad (IMO). Now, Google is providing a version of its model to the public.
Also: This facility can make GPT-5 a true game changer (if Openai is correct)
On Friday, Google launched Deep Think in the Gemini app for Google Ultra Subscribers, a premium subscription tier with a cost of $ 250 per year or $ 125 per year for the first three months. Although the model is a variation of one that has achieved the gold-foot standard in the IMO, it is sharp to everyday tasks. The internal assessment suggested that the model 2025 IMO reaches the bronze level on the benchmark.
How does this work
The better performance that solves complex problems is enabled by a parallel thinking technique, which allows the model to generate and process several ideas simultaneously, even in combination to find the best answers to different people.
Also: Google upgrade AI mode with canvas and 3 other new features – how to try them
Other factors contributing to high performance include an expanded conclusion time, also known as a thinking time, which allows to detect more options before reaching a north, and new reinforcement learning techniques that help the model become a better problem-secular.
According to Google, deep themes excel in the Itink Itress Development and Design, as seen in the above image, scientific and mathematical discovery and coding. These results are reflected in the performance of the state-of-the-art benchmark of Gemini 2.5 deep think, including the final examination of humanity, more than 100 subjects, such as mathematics, science, and an exam with multi-modal questions in humanities.
Google also shared that Gemini 2.5 Deep Think has shown better material safety and tone-objection than Gemini 2.5 Pro, with warnings that it has denied benign requests at a higher rate.
How to access
Google AI Ultra subscribers can use deep think in the Gemini app with a certain set of daily signals. To select the model, when selecting 2.5 Pro on the model selector, togle “deep think” to the Prompt bar. The company also shared that it is working to release deep thinks with devices and without equipment for a set of reliable examiners through Mithun API in the coming weeks.
Also: Openai teases the adjacent GPT-5 launch. What hope here
Gemini 2.5 deep think model that achieved the gold-foot standard will be shared with a small group of mathematicians and academics. The intention is that this model will be used to pursue their work, and it is expected that the lesson will provide feedback for improvement.
