Gemini 3 Deep Think Outsmarts All But Seven Humans in Programming
Google's Gemini AI Now Rivals Top Human Programmers
In a breakthrough that blurs the line between artificial and human intelligence, Google's Gemini 3 Deep Think has achieved what few thought possible: outperforming nearly every human programmer on Earth. With a Codeforces Elo rating of 3455, this AI now stands shoulder-to-shoulder with the world's coding elite - literally. Only seven living humans can currently claim superior algorithmic skills.
From Chessboards to Code Editors
The leap from its predecessor is staggering. Last year's strongest model scored 2727 - impressive, but still comfortably within human reach. This new version doesn't just compete; it dominates specialized domains:
- Scientific Peer Review: Detected subtle flaws in advanced physics papers that had already passed human review
- Mathematical Proofs: Cracked several stubborn problems related to the famous Erdős conjecture
- Engineering Design: Transforms napkin sketches into production-ready 3D models overnight
"We're seeing AI move from pattern recognition to genuine conceptual understanding," explains Dr. Elena Torres, a computational researcher at MIT not involved with the project. "When an algorithm starts catching mistakes that slip past PhD reviewers, we're entering uncharted territory."
Benchmarks Shattered
The numbers tell their own story:
- 48.4% on the grueling "Last Human Exam" (HLE)
- 84.6% accuracy on ARC-AGI-2 benchmark tests
- 10x faster at converting hand-drawn concepts to CAD models
Currently available to AI Ultra subscribers and select researchers via API, Gemini's upgrade arrives as tech giants race to develop reasoning systems that go beyond chatbots. Google appears determined to lead this next frontier.
Key Points:
- Only seven humans worldwide currently outperform Gemini in programming competitions
- Demonstrates unprecedented ability to critique advanced academic work
- Bridges creative design and technical execution through sketch-to-model conversion
- Marks significant progress toward artificial general intelligence (AGI)
