Asteromorf AI Posts Gold-Medal-Level Result at International Physics Olympiad
TECHWORLD ·
✦ AI Summary
Asteromorf said it entered its AI model in the actual International Physics Olympiad theory exam instead of using the conventional method of later solving publicly available problems.
The theory exam was held for 5 hours between 9 a.m. and 2 p.m. on July 8 under the same conditions as the actual participants, and the AI model scored 28.6 out of 30.
The organizing committee recognized the 28.6-point result as gold-medal level and awarded the AI model a diploma and medal listing it as a recipient.
Asteromorf said on the 31st that it had verified its AI model in a real competition environment, rather than relying on the conventional method of measuring performance by solving publicly available problems later. The Asteromorf AI model took part in the actual International Physics Olympiad exam.
The test was the theory exam at the 56th International Physics Olympiad (IPhO 2026). The International Physics Olympiad is an international physics competition for national teams of high school students, and consists of theory and experimental problem solving.
This year's competition was held last July in Bucaramanga, Colombia. For the event, Asteromorf sent an engineering team to the site and had the AI model solve problems for 5 hours, from 9 a.m. to 2 p.m. on July 8, the same time slot as the participants in the theory exam.
As a result, the Asteromorf AI model scored 28.6 out of 30. Asteromorf said the result was at the level corresponding to a gold medal at the International Physics Olympiad.
The grading of this answer was carried out in accordance with the competition's official procedures. Asteromorf said official International Physics Olympiad graders evaluated the submitted answers. The organizing committee then recognized the 28.6-point result as gold-medal level and awarded the AI model a diploma and medal listing it as a recipient.
The evaluation was carried out in line with the International Physics Olympiad's official grading system and the organizing committee's judgment, and the AI model was treated as a medal winner after being recognized at the gold-medal score threshold. The test conditions were also operated under the same settings as those for actual participants.
After the problem set was entered on site, the model generated answers within the time limit, and the same procedures were applied to grading and post-scoring adjustments. In other words, the entire test operation and follow-up process were carried out under the same conditions as for actual participants.
According to the company, the problem-solving process followed principles of not using online search, not using external AI model APIs, and not involving human intervention. From problem input through interpretation, reasoning, verification, and final answer generation, the AI system handled everything on its own. The system used for the exam was built by combining several open-source LLMs with Asteromorf's in-house AI models and components, and Amazon Web Services (AWS) cloud infrastructure was used for the computing environment.
To handle even non-text elements in the actual exam paper, the system applied multimodal technology. The elements it handled included the figures, graphs, and diagrams contained in the real test paper, and the problem text was processed by directly inputting the original exam paper without separate preprocessing. It also analyzed multiple types of information, including equations and images, at the same time, and the results of that simultaneous analysis were used to solve the physics problems.
Based on this, an Asteromorf official said the system was not dependent on the latest commercial models. The official added that the company had confirmed the possibility of solving advanced physics problems using only open-source models and its own AI systems, and emphasized the importance of the design and operational capabilities of a complex reasoning process system beyond the LLM's own performance.
Source: TECHWORLD · Kim Seung-gi
Original: https://www.epnc.co.kr/news/articleView.html?idxno=406283
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.