{"id":31066,"date":"2025-10-13T12:33:56","date_gmt":"2025-10-13T12:33:56","guid":{"rendered":"https:\/\/metaverseplanet.net\/blog\/?p=31066"},"modified":"2026-01-05T09:01:03","modified_gmt":"2026-01-05T09:01:03","slug":"samsung-developed-better-ai-than-models-10-thousand-times-larger","status":"publish","type":"post","link":"https:\/\/metaverseplanet.net\/blog\/samsung-developed-better-ai-than-models-10-thousand-times-larger\/","title":{"rendered":"Samsung Developed Better AI Than Models 10 Thousand Times Larger"},"content":{"rendered":"\n

Samsung<\/strong> has developed a groundbreaking artificial intelligence model<\/strong>. The model<\/strong>, named TRM<\/strong>, outperforms models 10 thousand times its size.<\/p>\n\n\n\n

Samsung<\/a><\/em><\/strong>, one of the world’s largest technology companies<\/strong>, has departments focused solely on developing groundbreaking new technologies<\/strong>. Naturally, AI-focused projects<\/strong> emerge from these areas. The latest news indicates that the company has developed an impressive artificial intelligence<\/a><\/em> model<\/strong>.<\/p>\n\n\n\n

Samsung researchers<\/strong>, led by Alexia Jolicoeur-Martineau, have succeeded in developing a small artificial intelligence model<\/strong> that can outperform<\/strong> models significantly larger than itself. This model<\/strong> has been named the Tiny Recursion Model (TRM)<\/strong>.<\/p>\n\n\n\n

\n\n\n\n

TRM Outperforms Models Like OpenAI o3 mini and Google Gemini 2.5 Pro<\/strong><\/h2>\n\n\n\n
\"\"<\/figure>\n\n\n\n

According to the information, TRM<\/strong> possesses a neural network<\/strong> with only 7 million parameters<\/strong>. This means it is a quite small artificial intelligence model<\/strong> compared to those currently on the market. However, despite its small size, TRM<\/strong> manages to outperform<\/strong> industry-leading models in the most challenging tests, such as OpenAI\u2019s<\/strong> o3 mini and Google’s<\/strong> most advanced model, Gemini 2.5 Pro<\/strong>. According to Samsung<\/strong>, it has succeeded in surpassing models 10 thousand times larger than itself.<\/p>\n\n\n\n

So, how does it achieve this? The company states that this success was achieved through a recursive reasoning approach<\/strong>. Martineau described TRM<\/strong> as a model that is “pre-trained from scratch, self-recursive, and updates its answers over time, capable of achieving a lot without breaking the bank.”<\/p>\n\n\n\n

The Hierarchical Reasoning Model (HRM)<\/strong> technique was developed and utilized in the model<\/strong>. This technique uses two collaborative networks, one operating at a higher frequency and the other at a lower frequency. Martineau and her team eliminated these elements and used a single double-layer model<\/strong>. This model<\/strong> continued to refine its predictions until it reached a sufficiently stable conclusion. Thus, despite being much smaller, it was able to demonstrate superior performance<\/strong>.<\/p>\n\n\n\n

You Might Also Like;<\/h3>\n\n\n