Researchers at Singapore-based AI company Sapient have announced a breakthrough in artificial intelligence with the launch of a new hierarchical reasoning model (HRM). Designed to mimic the way the human brain processes information across different timescales, the model is showing results that outperform well-known systems such as ChatGPT.
Unlike large language models that require billions of parameters and vast amounts of data, HRM is significantly smaller, with just 27 million parameters and only 1,000 training samples. Despite this, scientists claim it is more efficient and capable of handling complex reasoning tasks. The system uses a two-part design: a high-level module that manages slower, abstract planning and a low-level module that processes faster, detailed tasks.
The model was tested on the ARC-AGI benchmark, considered one of the most difficult tests to measure progress towards artificial general intelligence. In ARC-AGI-1, HRM scored 40.3 per cent, outperforming OpenAI’s 03-mini-high at 34.5 per cent, Anthropic Claude 3.7 at 21.2 per cent, and DeepSeek R1 at 15.8 per cent. In the tougher ARC-AGI-2 test, HRM reached 5 per cent, again surpassing the other models.
Scientists explained that while most advanced language models rely on chain-of-thought reasoning, this method has limitations such as brittle task breakdowns, high data needs, and slower responses. HRM instead applies sequential reasoning in a single forward pass, paired with an iterative refinement method. This means it begins with an approximate answer and refines it through short bursts of reasoning until the outcome is accurate.
In trials, HRM was able to solve Sudoku puzzles and navigate complex mazes, tasks that conventional language models usually struggle with. Researchers highlighted that this shows HRM’s strength in solving structured and logical problems.
The study has been published in the arXiv database but has not yet been peer reviewed. Independent researchers recreated the experiments and confirmed the performance results, though they suggested that a less-documented refinement step in training might explain the strong numbers, rather than the hierarchical design alone.
Also read: Viksit Workforce for a Viksit Bharat
Do Follow: The Mainstream formerly known as CIO News LinkedIn Account | The Mainstream formerly known as CIO News Facebook | The Mainstream formerly known as CIO News Youtube | The Mainstream formerly known as CIO News Twitter |The Mainstream formerly known as CIO News Whatsapp Channel | The Mainstream formerly known as CIO News Instagram
About us:
The Mainstream formerly known as CIO News is a premier platform dedicated to delivering latest news, updates, and insights from the tech industry. With its strong foundation of intellectual property and thought leadership, the platform is well-positioned to stay ahead of the curve and lead conversations about how technology shapes our world. From its early days as CIO News to its rebranding as The Mainstream on November 28, 2024, it has been expanding its global reach, targeting key markets in the Middle East & Africa, ASEAN, the USA, and the UK. The Mainstream is a vision to put technology at the center of every conversation, inspiring professionals and organizations to embrace the future of tech.
