Publication
Where do small open language models break on olympiad mathematics? Domain, difficulty and model size on Omni-MATH - Central Asian Scientific Journal, 2026
2026
Open
Public profile
Amirzhan Idrissov
@abdweeq
High School Student · Kokshetau BIL
I’m a high school student at Kokshetau BIL interested in olympiad mathematics, AI/ML, research, competitive programming, and mathematical visualization. I work mainly with Python and Manim and aim to grow toward machine learning engineering.
Astana
Featured
A few things that best show this member's work and contribution.
How I can help
Mathematical problem solving, olympiad math experience, ideas, mathematical visualization with Manim, and collaboration on projects and competitions.
Looking to connect
I’m looking for mentors, research and project collaborators, opportunities to join DSML Young, participate in AI/ML competitions, and learn machine learning through real projects.
Research and publications
A compact preview of this member's public research layer.
I am a school student in Kazakhstan interested in how language models solve mathematical problems, especially in algebra and geometry. My first research paper compared Qwen2.5-Instruct models with 1.5B, 3B and 7B parameters on 287 olympiad problems from Omni-MATH. I examined how performance varies across mathematical domains and difficulty levels, and whether missing diagrams explain geometry errors. Larger models improved much more in algebra than in geometry, and removing problems with missing diagrams did not close the gap. I would like to continue exploring mathematical reasoning, model evaluation and error analysis.
Language modelsmathematical reasoningLLM evaluationolympiad mathematicserror analysis