Google Unveils Gemini 3 Deep Think: New Benchmark Leader Achieves 84.6% on ARC-AGI-2
Google has released an updated version of Gemini 3 Deep Think, which according to benchmark results, represents the most advanced publicly available neural network model to date.
The new Deep Think model has achieved record-breaking performance metrics:
• 84.6% on the ARC-AGI-2 benchmark
• 48.4% on Humanity's Last Exam
• Gold medal performance at the International Mathematical Olympiad
Building upon this foundation, Google DeepMind has developed Aletheia, an AI agent that ranks among the world's leading AI mathematicians. The bot has successfully solved four previously unsolved mathematical problems, demonstrating unprecedented capabilities in advanced mathematical reasoning.
This release marks a significant milestone in AI development, particularly in domains requiring deep analytical thinking and complex problem-solving capabilities.
Sources:
https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-deep-think/
https://deepmind.google/blog/accelerating-mathematical-and-scientific-discovery-with-gemini-deep-think/
🔔 Stay tuned and subscribe →
Related news
Try these AI tools
Compare ChatGPT, Claude, Gemini, and Perplexity answers in one search. Find the best AI response fas...
Google DeepMind: pioneering AI for science and society, with models like Gemini and AlphaFold.
Interact with multiple AI chatbots (ChatGPT, Gemini, Perplexity, Claude, ), including Threads by Ins...
