Story perspectives

Google's Gemini 3 Solves Decade-Old Math Conjecture

2/14/2026

Summary

Story summary

Google's Gemini 3 Deep Think reasoning model has been upgraded and launched for scientific and engineering challenges.
The update is available to Google AI Ultra subscribers and select researchers via API.
The model improves performance in mathematics, physics, programming.
Benchmarks show 48.4% on Humanity's Last Exam, 84.6% on ARC-AGI-2, and a Codeforces Elo rating of 3455.
It has solved 18 previously unsolved research problems, including disproving a decade-old mathematical conjecture.

History

Historical Context

In March 2016, DeepMind's AlphaGo made history by defeating Go champion Lee Sedol 4-1, showcasing AI's advanced problem-solving capabilities. This victory, unexpected by many experts, marked a turning point in AI research, as AlphaGo utilized deep learning and Monte Carlo tree search techniques. The success of AlphaGo has since influenced significant investments in AI, particularly in China, highlighting its transformative potential in technology and research.

1 of 2

Story summary

Google's Gemini 3 Deep Think reasoning model has been upgraded and launched for scientific and engineering challenges.
The update is available to Google AI Ultra subscribers and select researchers via API.
The model improves performance in mathematics, physics, programming.
Benchmarks show 48.4% on Humanity's Last Exam, 84.6% on ARC-AGI-2, and a Codeforces Elo rating of 3455.
It has solved 18 previously unsolved research problems, including disproving a decade-old mathematical conjecture.

1 / 2