Alibaba launches its largest AI model: Qwen3-Max

Alibaba, the Chinese company, has recently unveiled its most capable artificial intelligence (AI) model to date, called "Qwen3-Max", which contains more than 1 trillion parameters, trained on 36 trillion tokens. 

According to a recent blog by Qwen, a series of large language models (LLMs) developed by Alibaba Cloud, the AI model achieves 'state-of-the-art' results across benchmarks in knowledge, reasoning, coding, instruction following and multilingual understanding. A preview version of Qwen3-Max-Instruct, one of the model's variants, had already placed third on the Text Arena leaderboard, surpassing GPT-5-Chat in certain tests. 

On SWE-Bench Verified, a benchmark for solving real-world coding tasks, it achieved a score of 69.6, while on Tau2-Bench, a test of how well AI models use tools, Qwen3-Max-Instruct achieved a breakthrough score of 74.8, surpassing both Claude Opus 4 and DeepSeek V3.1.

Another version of the system, Qwen3-Max-Thinking, remains in training but has already demonstrated exceptional results, including perfect scores on complex mathematical reasoning tests such as AIME 25 and HMMT. 

Qwen3-Max is now available for use through Alibaba Cloud's Model Studio API and via Qwen Chat.

The announcement comes as Alibaba seeks to increase its AI credentials alongside its core e-commerce operations. Earlier this year, the company pledged to invest 380 billion yuan ($53.4 billion) in AI-related infrastructure over the next three years.

In addition to Qwen3-Max, Alibaba also introduced Qwen3-Omni, a multimodal model designed for virtual and augmented reality applications, including smart glasses and intelligent vehicle cockpits.