Alibaba Launches Qwen-3 Max

Alibaba has unveiled Qwen-3 Max, its most advanced large language model (LLM) to date, aiming to compete with leading models like OpenAI’s GPT-5, Google’s Gemini 2.5 Pro, and Claude’s Opus 4. This is the first model in the Qwen series to exceed one trillion parameters, having been trained on 36 trillion tokens. With a context window of one million tokens, Qwen-3 Max can process entire codebases and lengthy documents efficiently.

The model brings enhancements in reasoning, instruction following, multilingual capabilities, and domain-specific knowledge, including improved performance in math, coding, logic, and science tasks. It also offers better comprehension in both English and Chinese, with fewer hallucinations and more accurate open-ended responses.

Qwen-3 Max ranks third on LMArena’s text leaderboard, ahead of the standard GPT-5, and scores 69.6 on the SWE-Bench Verified coding benchmark. It also outperforms Claude Opus 4 in Tau2-Bench with a score of 74.8. Alibaba also teased Qwen-3 Max Thinking, still in training, which reportedly scores perfectly on some reasoning benchmarks.

You can try Qwen-3 Max for free via the Qwen app or website. On mobile, it’s the default model—switch it manually if not pre-selected.