Google has updated the reasoning mode of Gemini 3 Deep Think. The tool is positioned as a solution for complex tasks in science and engineering.
In tests, the model outperformed OpenAI’s GPT-5.2 and Anthropic’s Claude Opus 4.6, including in ARC-AGI-2 with visual puzzles, MMMU-Pro for evaluating multimodal capabilities, Elo 3455, and the “Last Exam of Humanity.”
“We updated Gemini 3 Deep Think in close collaboration with scientists and researchers to tackle complex scientific challenges—where tasks often lack clear boundaries or a single correct solution, and data is incomplete,” the company blog states.
Gemini 3 Deep Think demonstrates advanced results in mathematics and programming, and performs “excellently” in natural sciences, including chemistry and physics. The updated mode solves problems at the level of gold medalists in international olympiads.
In the CMT-Benchmark, the model scored 50.5%, confirming deep knowledge in theoretical physics.
“Beyond advanced metrics, Deep Think is geared towards practical application: it helps researchers interpret complex data and engineers model physical systems through code,” Google noted.
The new Deep Think is available in the Gemini app for Google AI Ultra subscribers and Gemini API for select developers.
AI Mathematician from DeepMind
Google’s DeepMind division introduced the AI agent Aletheia. The model set a new record in the IMO-ProofBench Advanced benchmark, solving 91.9% of tasks. The test is considered one of the most challenging in mathematics.
The neural network is built on the Gemini Deep Think platform. The system is equipped with a verification module: it identifies errors in draft solutions and initiates an iterative process of refinement.
A key feature of the agent is its ability to recognize the impossibility of solving a problem, significantly saving researchers’ time.
Aletheia uses Google Search to navigate complex scientific materials, preventing the use of false links and computational errors when working with scientific materials.
Among the model’s achievements:
- complete generation of a scientific paper calculating structural constants in arithmetic geometry;
- collaborative proof of estimates for systems of interacting particles (independent sets) with a human;
- autonomous solution of four problems from the Erdős list, one of which was previously considered open.
DeepMind emphasized that Aletheia’s success confirms the relevance of scaling laws: in proof-based mathematics, quality continues to improve through the effective application of agents.
Breakthrough in Medicine
DeepMind’s subsidiary, Isomorphic Labs, introduced the IsoDDE engine for drug development. In complex tests, the innovation outperformed AlphaFold 3 in prediction accuracy by a factor of two.
The latter was a major breakthrough as it could predict the three-dimensional structures of proteins and their interactions with molecules. IsoDDE, however, demonstrates an entirely new level:
- the model predicts binding strength (affinity) more accurately than traditional methods;
- the engine can identify hidden structures (“pockets”) in proteins where drugs can bind;
- it supports a wide range of complex molecules, including antibodies and large biological structures.
“IsoDDE offers a scalable framework for AI-driven drug design, providing the prediction accuracy necessary to work with new biological systems with unprecedented reliability,” the company blog states.
Back in July 2022, the AlphaFold algorithm predicted the structures of over 200 million proteins. This encompasses nearly all known compounds found in plants, bacteria, and animals.