Major Upgrade! Gemini 3 Deep Think: Reasoning Dramatically Enhanced

AI

📝 Key Points of This Article

• Google has announced a major upgrade to “Gemini 3 Deep Think,” its top-tier AI model available to “Google AI Ultra” members.
• The model’s reasoning capabilities in highly specialized fields such as science, research, and engineering have been dramatically enhanced.
• Its defining feature is the ability to derive solutions for complex challenges that have traditionally been difficult for AI—such as problems with no single correct answer or those involving incomplete data—by combining deep scientific knowledge with a practical engineering perspective.

On Thursday, February 12, 2026, Google announced a major upgrade to “Gemini 3 Deep Think,” its top-tier AI model currently being rolled out for “Google AI Ultra” members within the “Google One” “Google AI” subscription plan.

With this major upgrade, Gemini 3 Deep Think’s reasoning capabilities in highly specialized fields such as science, research, and engineering will be dramatically strengthened. Its most significant feature is the ability to derive solutions for complex challenges that have traditionally been difficult for AI—such as problems where there is no single right answer or where only incomplete data exists—by combining deep scientific knowledge with a practical engineering perspective.

The performance of the upgraded Gemini 3 Deep Think has already been demonstrated in academia and competitive fields, achieving scores equivalent to a gold medal in the written exams of the 2025 International Mathematical, Physics, and Chemistry Olympiads. Furthermore, it reached a top-tier Elo rating of 3,455 on the programming contest site “Codeforces” and achieved world-class results on benchmarks measuring advanced reasoning, such as “ARC-AGI-2” and “Humanity’s Last Exam.”

These capabilities go beyond mere improvements in computing speed and are already yielding concrete results in real-world research and development. For instance, at Rutgers University, the AI identified subtle logical flaws in mathematical papers that human experts had overlooked, while a study at Duke University successfully optimized the manufacturing process for semiconductor materials. Additionally, it offers practical utility that accelerates the manufacturing process itself, such as generating 3D printing data from hand-drawn sketches.

Gemini 3 Deep Think is available to “Google AI Ultra” members through the Gemini app, and an early access program via the Gemini API has also been launched for researchers and enterprises.

Source: Google