Alibaba launches Qwen3-Max AI model to challenge GPT-5 Pro

Chinese tech giant Alibaba unveils Qwen3-Max, a new AI model designed to compete with leading systems like GPT-5 Pro and Grok 4.
At Alibaba's 2025 Apsara Conference on September 24, the company announced Qwen3-Max, its largest artificial intelligence model. As Alibaba research scientist Binyuan Hui wrote on X, the new model introduces improved coding and agent capabilities.
According to the announcement, Qwen3-Max has over 1 trillion parameters and was pretrained on 36 trillion tokens, making it highly capable at understanding, reasoning, and generating language. In its current preview version, the model ranks third on the Text Arena leaderboard, surpassing OpenAI’s GPT-5-Chat, while Google’s Gemini 2.0 Flash and Meta’s LLaMA-4 hold first and second place.
The launch shows Alibaba is stepping up its efforts in the global AI race. The company follows other Chinese firms like DeepSeek, which recently trained its R1 AI model for just $294,000 - much less than the $100 million often spent on similar projects in the West.
Alibaba also introduced Qwen3-Omni, a system for virtual and augmented reality, including smart glasses and in-car “intelligent cockpit” systems. Earlier this year, the company announced a 380 billion yuan (~$53.4 million) investment in AI infrastructure over three years, with plans to spend even more. The Qwen 3 family was first released in April.
Qwen3-Max-Instruct is now available in Qwen Chat, allowing anyone to interact directly with the AI. Its API (model name: qwen3-max) can also be accessed through an Alibaba Cloud account, letting developers try out the model and use it in different applications.
The AI model competition is increasingly global, with the U.S. and China leading in development and investment. In 2024, the U.S. produced 40 notable AI models, while China produced 15, but Chinese models are rapidly closing the performance gap on key benchmarks, according to The 2025 AI Index Report by Stanford University.
