• by Laurie Sullivan , February 15, 2026

Google DeepMind released an upgrade to Gemini 3 Deep Think that specializes in reasoning.

It is designed to tackle complex math, science, and logic challenges, but advertisers and agencies can use it to solve questions related to media buying or campaign research, as well as build agents without code.

Advanced reasoning is used to create agentic models that automate the process of turning creative briefs into campaign recommendations.

For example, Gemini 3 Deep Think could analyze incomplete data and combine other types of data related to the brand to gain insights.

The model can analyze experimental data, find patterns, and identify flaws even if context and content are lacking.

Lisa Carbone, a mathematician at Rutgers University, worked on a paper with colleagues that took several years.

Before sending it out to the journal, she put it through Gemini to fact-check the content for verification.

“It came back with, no, that’s not correct,” Carbone explained in a video.

The model told her the findings were mathematically incorrect as stated, and provided three irrefutable reasons around the arguments.

One statement was incompatible. When she refuted the findings, the model did not attempt to appease her. It took her time to understand because the findings were outside of her thought process.

“The model’s reasoning was completely correct,” she said.

The research papers, she said, are at the forefront of the subject, so very little context or training data exists that the models could have been trained on.

The model uses inference-time compute -- computational power that an AI model uses while it generates an answer, rather than the power used to train it.

It has the knowledge to simulate different solutions and check its logic. This allows it to handle real-world research data.

Google also is working to bring Deep Think to researchers and practitioners beginning with surfaces such as the Gemini API. As of this month, the API provides specific agentic capabilities and controls to solve complex problems.

Agencies can use the Gemini API to build "no-code" agents to automate campaigns, from research to reporting.

The advancement comes in what is called the “thinking-level parameter,” which controls Gemini 3 models and allows developers to adjust its reasoning on a per-request basis.
High levels are for deep planning and complex logic, while low levels prioritize speed and high throughput, according to several reports.

Advertisers can use Gemini 3’s computational technology to tailor their approach to the complexity of each marketing task.

Google AI Ultra subscribers gained access to the updated Deep Think mode last week in the Gemini app.