In a reasoning test using Arena-Hard, Qwen 2.5-Max achieved 89.4% accuracy, and the result was higher than DeepSeek R1 and when tested on other benchmarks of coding and scientific reasoning, Qwen 2.5 ...