Singaporean Startup Sapient Intelligence Makes Leap with New AI Architecture
singapore, vrijdag, 1 augustus 2025.
Sapient Intelligence, a young startup from Singapore, has developed a promising approach to General Artificial Intelligence (AGI). The company uses a hierarchical architecture, known as HRM, which outperforms larger models in complex reasoning tasks. With just 27 million parameters and approximately 1,000 training examples, the HRM model succeeds in tasks where other models fail, such as solving complex Sudoku puzzles and finding the optimal route in 30x30 mazes. This offers new perspectives in areas where accuracy and interpretability are essential, such as medical diagnostics and climate prediction.
A New Approach to General Artificial Intelligence
Sapient Intelligence, a young startup from Singapore, has developed a promising approach to General Artificial Intelligence (AGI). The company uses a hierarchical architecture, known as HRM (Hierarchical Reasoning Model), which outperforms larger models in complex reasoning tasks. With just 27 million parameters and approximately 1,000 training examples, the HRM model succeeds in tasks where other models fail, such as solving complex Sudoku puzzles and finding the optimal route in 30x30 mazes [1][2].
Efficient and Accurate Reasoning
The HRM model employs a two-layer hierarchical structure: a high-level recurrent network that makes abstract and slow plans, and a lower level that performs fast and detailed tasks. This approach provides a significant improvement in efficiency and accuracy, achieving a performance of 5% on version 2 of ARC-AGI, one of the most stringent benchmarks for inductive intelligence [1][2].
Applications in Critical Areas
The efficiency and low data dependency of the HRM model offer potential in domains where accuracy and interpretability are essential, such as medical diagnostics and climate prediction. The model can, for example, be used for diagnosing rare conditions and seasonal climate predictions with 97% accuracy [1][2].
Vision of the Researchers
Guan Wang, founder and CEO of Sapient Intelligence, emphasises the company’s goal: ‘AGI is really about giving human-level intellect to machines, and ultimately more. Our model thinks and reasons like a person, and is not limited to analysing probabilities. We believe it will reach and surpass human intelligence, and that is when the AGI conversation will truly begin [1][2].’
Impact and Future Perspectives
The HRM architecture is inspired by how the human brain processes information, focusing on hierarchical structure and multi-scale time processing. This offers new possibilities in various sectors, including medical diagnostics, financial analysis, and security checks. Applying this model in these areas can lead to significantly better results and higher efficiency [1][2].