[R] Analysis of 350+ ML competitions in 2025
Analysis of 400 competitions shows Qwen models dominate text tasks, while one team spent $60k on 512 H100s.
Deep Dive
MLContests.com analyzed over 400 machine learning competitions from 2025. Key findings: Qwen2.5/3 models dominated text/reasoning wins, while XGBoost still led tabular data. Compute budgets soared, with one NVIDIA team using $60k of H100 GPUs. For the first time, Transformer-based models won more vision competitions than CNNs. The report details winning solutions, tools like vLLM, and Python package trends, providing a benchmark for the competitive ML field.
Why It Matters
This report is a crucial benchmark for data scientists, revealing the most effective models, tools, and compute strategies winning real-world challenges.