VERSES® Genius™ Outperforms DeepSeek R1 Model in Code-Breaking “Mastermind” Challenge

Core Insights - VERSES AI Inc. announced that its flagship product, Genius, outperformed both OpenAI's o1-preview model and China's DeepSeek's R1 model in a code-breaking game called Mastermind, showcasing its advanced capabilities in multi-step reasoning and efficiency [1][4][5] Performance Comparison - Genius achieved a 100% success rate in cracking the Mastermind code, while DeepSeek's R1 only succeeded 45% of the time [2][5] - The total compute time for Genius to complete 100 games was 5 minutes and 18 seconds, averaging 3.1 seconds per game, compared to DeepSeek's 26 hours, averaging 334 seconds per game [2][5] - The estimated total cost for Genius to solve 100 games was $0.05, whereas DeepSeek's R1 incurred a cost of$38.94 [2][5] Technological Advantages - Genius demonstrated a performance that was 245 times faster and 779 times cheaper than DeepSeek's R1 model [3][4] - The technology behind Genius leverages a Bayesian approach and Active Inference, allowing it to excel in dynamic real-world scenarios [4] Market Implications - The results of the Mastermind challenge suggest that domain-specific models like Genius could be crucial for enhancing the capabilities of general-purpose large language models (LLMs) [4] - The CEO of VERSES indicated that the accuracy and reliability of AI agents are key to unlocking broader adoption across major corporations [4]