Google has rolled out a major upgrade to Gemini 3 Deep Think, a specialized reasoning mode designed to handle complex scientific, mathematical and engineering problems that exceed the capabilities of standard generative AI. The update marks a shift in Google’s AI strategy away from conversational novelty and toward applied reasoning for real world use cases. The tech mogul announced:
Today, we’re releasing a major upgrade to Gemini 3 Deep Think, our specialized reasoning mode, built to push the frontier of intelligence and solve modern challenges across science, research, and engineering.
Deep Think operates as an advanced reasoning layer within the broader Gemini 3 architecture, combining text, visual understanding and logical inference with iterative hypothesis testing. Rather than generating quick answers, the system allocates additional internal reasoning time, explores multiple solution paths in parallel and refines conclusions before responding. This approach allows it to work through messy datasets, multi step calculations and ambiguous technical problems that typically require expert human judgment.
According to Google, here is how Gemini will improve:
The updated Deep Think mode continues to push the frontiers of intelligence, reaching new heights across the most rigorous academic benchmarks, including:
- Setting a new standard (48.4%, without tools) on Humanity’s Last Exam, a benchmark designed to test the limits of modern frontier models
- Achieving an unprecedented 84.6% on ARC-AGI-2, verified by the ARC Prize Foundation
- Attaining a staggering Elo of 3455 on Codeforces, a benchmark consisting of competitive programming challenges
- Reaching gold-medal level performance on the International Math Olympiad 2025
The upgraded Deep Think mode shows significant gains on demanding benchmarks covering mathematics, logic and scientific reasoning. These include strong performance on abstract reasoning tests and competitive results in physics and chemistry style problem sets. The improvements reflect a broader industry trend toward higher order reasoning models that prioritize depth and reliability over speed.
The update is currently available to Google AI Ultra subscribers, with access through the Gemini app and limited early access via the Gemini API for researchers and enterprise users. By exposing Deep Think through programmable interfaces, Google is signaling that this is not just a research demo but a tool meant to integrate into professional workflows across academia, industrial R&D and advanced engineering teams.
Experts note that systems like Deep Think could materially change how research is conducted. Instead of replacing human expertise, the model acts as a reasoning partner, helping teams test assumptions, explore alternative hypotheses and validate complex ideas faster. In environments where problems do not have clean inputs or obvious solutions, that kind of support can accelerate discovery without compromising rigor.
Google AI Ultra subscribers will be able to access the updated Deep Think mode in the Gemini app. Scientists, engineers and enterprises can also now express interest in an early access program to test Deep Think via the Gemini API. Given how OpenAI is expanding GPT’s services to benefit the scientific community, this can be taken as Google’s response to the ongoing AI wars.
