Google has announced a major upgrade to its Gemini 3 Deep Think, its specialised reasoning mode designed to go beyond the conversational capabilities of standard AI models. The update is intended to address difficult problems in science, research and engineering. It is designed to handle situations where “data is often messy or incomplete,” Google said in a blog post.
Sundar Pichai Highlights Major Upgrade
In a post on X, Sundar Pichai, the CEO of Google and Alphabet, said, “Gemini 3 Deep Think is getting a significant upgrade. We've refined Deep Think in close partnership with scientists and researchers to tackle tough, real-world challenges. And it's pushing the frontier across the most challenging benchmarks, achieving an unprecedented 84.6% on ARC-AGI-2. It also sets a new standard on Humanity's Last Exam - 48.4% without tools.”
Gemini 3 Deep Think is getting a significant upgrade. We've refined Deep Think in close partnership with scientists and researchers to tackle tough, real-world challenges.
— Sundar Pichai (@sundarpichai) February 12, 2026
And it's pushing the frontier across the most challenging benchmarks, achieving an unprecedented 84.6% on… pic.twitter.com/5503F4FKcD
Pichai added in a subsequent post, “The updated Gemini 3 Deep Think mode is now available for Ultra subscribers in the Gemini app. We're also, for the first time, making Deep Think available via the Gemini API to select researchers & enterprises through an early access program.”
The updated Gemini 3 Deep Think mode is now available for Ultra subscribers in the Gemini app.
— Sundar Pichai (@sundarpichai) February 12, 2026
We're also, for the first time, making Deep Think available via the Gemini API to select researchers & enterprises through an early access program.
Read more: https://t.co/I6zKGf0VjQ
Also Read: Anthropic Finalises $30 Billion Funding at $380 Billion Value
Setting New Benchmarks
According to Google, the updated Deep Think mode has already shown strong performance in competitive reasoning tasks. In 2025, specialised versions of the system achieved gold-medal level results in mathematics and programming world championships. The new model continues this trend by setting benchmarks across multiple academic tests. It scored 48.4% on Humanity's Last Exam, a test designed to challenge frontier AI models. It achieved 84.6% on ARC-AGI-2, as verified by the ARC Prize Foundation. It also attained an Elo rating of 3455 on Codeforces and reached gold-medal standard in the International Math Olympiad 2025.
Expanding Across Disciplines
Beyond mathematics and coding, Gemini 3 Deep Think now demonstrates high-level performance in physics and chemistry. It achieved gold-medal standard results on the written sections of the International Physics Olympiad 2025 and the International Chemistry Olympiad 2025. The AI also displayed proficiency in advanced theoretical physics, scoring 50.5% on the CMT-Benchmark.
Practical Engineering Applications
Google said Deep Think is not only about academic performance. The system is built to support real-world work, helping researchers analyse complex data and engineers model physical systems. By combining scientific knowledge with practical engineering, Deep Think is designed to assist professionals in addressing modern technological challenges.
Availability
The updated Deep Think mode is now accessible to Google AI Ultra subscribers through the Gemini app. Google is also offering early access to the Gemini API, allowing select researchers, engineers and enterprises to test the capabilities of Deep Think.
Essential Business Intelligence, Continuous LIVE TV, Sharp Market Insights, Practical Personal Finance Advice and Latest Stories — On NDTV Profit.
