Alibaba’s AI aces top global maths contests, challenging OpenAI’s dominance
Qwen3-Max-Thinking, the latest version of Alibaba’s Qwen3 model family, achieved perfect scores in two prestigious US maths competitions
Chinese technology giant Alibaba Group Holding on Monday unveiled an upgraded artificial intelligence reasoning model that excelled in leading global maths competitions, matching the performance of US rival OpenAI’s most advanced model.
Qwen3-Max-Thinking, the latest version of Alibaba’s Qwen3 model family, achieved 100 per cent accuracy “in challenging reasoning benchmarks like American Invitational Mathematics Examination (AIME) 2025 and Harvard-MIT Mathematics Tournament (HMMT)”, the Qwen team said, making it the first Chinese AI reasoning model to score 100 per cent in the games. Alibaba owns the Post.
AIME and HMMT are among the most challenging and prestigious global competitions that test problem-solving in areas such as arithmetic, algebra, number theory and probability.
A high score in mathematical reasoning tests was essential for evaluating an AI model’s reasoning and problem-solving skills, according to Intuition Labs, a San Jose-based AI software firm. These contests represented a key frontier in AI advancements in the race to build stronger reasoning models, it added.
10:41
How Hangzhou’s ‘Six Little Dragons’ built a new Chinese tech hub
Other models that have achieved perfect scores in these competitions include GPT-5 Pro, OpenAI’s most advanced model released in August, according to self-reported results from the Microsoft-backed company.
Qwen3-Max-Thinking was developed using Qwen3-Max. This model, the largest variant from Alibaba’s AI and cloud services unit, has over 1 trillion parameters and was unveiled in late September. The original Qwen3 was released in April.