0:00:01 Benchmark Performance: Gro 3 and Gro 3 mini surpass state-of-the-art models like Gemini 2 Deep, Seek V3, Claude 3.5 Sonnet, and GPT-4o across various benchmarks, even when tested on new benchmarks.
0:01:17 Evaluation Categories: Gro 3 was evaluated in general mathematics, STEM knowledge, and computer science coding using tests like the American Invitational Mathematics Examination (AIME). Gro 3 demonstrated exceptional performance.
0:02:16 Chatbot Arena: Gro 3 excels in real-time usefulness, ranking #1 in the Chatbot Arena, a blind test where users choose the better response between two AI models without knowing which model generated it.
0:03:00 Blind Test: An early version of Grock 3, codenamed "Chocolate", already reached an ELO score of 1400 in a blind test on the Chatbot Arena, outperforming other models across multiple categories.
0:04:41 Continuous Improvement: Gro 3 is continuously improving, with users potentially noticing enhancements daily.
0:04:46 Reasoning Capabilities: Gro 3's reasoning models surpass even the recently released Gro 3 mini. They are trained to think for extended periods, leading to more accurate and complex problem-solving. More compute time enhances results.
0:06:51 Benchmarks: Gro 3 performs well in mathematics (high school competition problems), science (PhD level science questions), and coding (competitive coding and LeetCode problems).
0:08:27 Generalization: Gro 3 shows true generalization, excelling in the new AIME 2025 competition.
0:10:27 Reasoning Demonstration: Gro 3's reasoning capabilities are showcased by generating code for an animated 3D plot of a launch from Earth to Mars and back.
0:13:37 Deep Search: A new product called Deep Search, is introduced. This is the first generation of Gro agents that help engineers, researchers, and scientists do coding but also help everyone to answer questions that you have dayto day.
0:13:57 Deep Search Functionality: Deep Search is a next-generation search engine which understands user intent, considers relevant facts, and cross-validates sources, potentially saving hours of research time. It includes transparent chain of thought capabilities for better user understanding and trust.
0:17:33 Rollout and Access: Gro 3 and Deep Search will be accessible via gro.com. The Super Gro subscription will offer access to the most advanced capabilities and early access to new features. The Grock app is also available in the iOS app store. The web version at gro.com will be the latest and most advanced.