Home
New Gemini 3 Deep Think Released, Tops Programming and Research Rankings — Only 7 Can Beat It
Google's Gemini3Deep Think received a significant update today. Designed for deep reasoning, this model not only achieves champion‑level performance in programming but also sets new records in scientific research and hardcore engineering, marking a leap forward in AI reasoning capabilities.

Gemini3Deep Think's standout achievements:
Programming mastery: It achieved an impressive Elo score of 3455 on the Codeforces algorithm competition platform. That places it among top human competitors, with only seven humans worldwide capable of beating it. For context, the strongest o3 model from a year ago scored 2727.
Powerful brain for scientific research: The model demonstrates rigorous logical reasoning, catching subtle errors in a high‑level physics and mathematics paper that even human peer reviewers overlooked. It also succeeded in proving several challenging problems related to the "Erdős conjecture."
From sketch to physical product: In engineering, it can directly analyze a hand‑drawn sketch and render it into a high‑fidelity 3D model file (like a notebook stand), boosting the efficiency of physical component modeling by tenfold.
Comprehensive breakthroughs in benchmark tests: It scored 48.4% on the "Last Human Exam" (HLE) and achieved an impressive 84.6% accuracy on the ARC‑AGI‑2 benchmark, leading the rankings.
Currently, Google has made the new version available to AI Ultra subscribers and, for the first time, provided API access to select researchers and companies. This major upgrade to Gemini is seen as a strong response to competitors' reasoning models.
Related article
Apple iOS 27 Leaks: Siri Integrates Google Gemini for Local AI, With Some Requests Routing to Google Cloud
Recent reports reveal the depth of Apple’s partnership with Google in artificial intelligence. The upcoming iOS 27 will prioritize privacy by focusing on local edge AI processing. To boost performance without compromising user data, Apple is employin
Anthropic Eyes Record 2026 IPO as Six-Fold Revenue Growth Nears Musk's Market Cap
Despite not yet going public, AI startup Anthropic has already sparked intense activity in prediction markets. Latest Polymarket data shows investors betting Anthropic will surpass Elon Musk’s SpaceX to claim the title of the world’s largest IPO in 2
U.S. Military Contractors Phase Out Claude Model Amid Tighter Restrictions and Conflicts
U.S. defense technology firms face supply chain disruptions due to conflicting regulatory policies, as reported by TechCrunch. While the U.S. military continues to utilize Anthropic’s Claude for operational support, overlapping bans from the Trump ad
Related Special Topic Recommendations
Comments (1)
0/500
Google's Gemini3Deep Think received a significant update today. Designed for deep reasoning, this model not only achieves champion‑level performance in programming but also sets new records in scientific research and hardcore engineering, marking a leap forward in AI reasoning capabilities.

Gemini3Deep Think's standout achievements:
Programming mastery: It achieved an impressive Elo score of 3455 on the Codeforces algorithm competition platform. That places it among top human competitors, with only seven humans worldwide capable of beating it. For context, the strongest o3 model from a year ago scored 2727.
Powerful brain for scientific research: The model demonstrates rigorous logical reasoning, catching subtle errors in a high‑level physics and mathematics paper that even human peer reviewers overlooked. It also succeeded in proving several challenging problems related to the "Erdős conjecture."
From sketch to physical product: In engineering, it can directly analyze a hand‑drawn sketch and render it into a high‑fidelity 3D model file (like a notebook stand), boosting the efficiency of physical component modeling by tenfold.
Comprehensive breakthroughs in benchmark tests: It scored 48.4% on the "Last Human Exam" (HLE) and achieved an impressive 84.6% accuracy on the ARC‑AGI‑2 benchmark, leading the rankings.
Currently, Google has made the new version available to AI Ultra subscribers and, for the first time, provided API access to select researchers and companies. This major upgrade to Gemini is seen as a strong response to competitors' reasoning models.
Apple iOS 27 Leaks: Siri Integrates Google Gemini for Local AI, With Some Requests Routing to Google Cloud
Recent reports reveal the depth of Apple’s partnership with Google in artificial intelligence. The upcoming iOS 27 will prioritize privacy by focusing on local edge AI processing. To boost performance without compromising user data, Apple is employin
Anthropic Eyes Record 2026 IPO as Six-Fold Revenue Growth Nears Musk's Market Cap
Despite not yet going public, AI startup Anthropic has already sparked intense activity in prediction markets. Latest Polymarket data shows investors betting Anthropic will surpass Elon Musk’s SpaceX to claim the title of the world’s largest IPO in 2
U.S. Military Contractors Phase Out Claude Model Amid Tighter Restrictions and Conflicts
U.S. defense technology firms face supply chain disruptions due to conflicting regulatory policies, as reported by TechCrunch. While the U.S. military continues to utilize Anthropic’s Claude for operational support, overlapping bans from the Trump ad











