Alibaba has released its strongest AI yet on September 24, 2025. According to the Alibaba Cloud website, the open-source Qwen series saw 7 million downloads by May 2024 (just after launching in April 2023). Notably, Qwen is now ranked third globally on the LMArena text leaderboard. To best serve its users, this Chinese AI has made significant improvements to its new version. So, the new model is all set to compete with AI giants like OpenAI’s GPT-5, Google’s Gemini 2.5 Pro, and Anthropic’s Claude Opus 4. How is the new version different? How does it rank compared to others? Learn more.
What Is Qwen-3 Max?
- Qwen-3 Max is part of Alibaba’s Qwen series of AI models, also known as the Qwen family.
- This is the first Qwen model to surpass the 1 trillion parameters (these parameters are the connections in the brain of the AI. More parameters mean a more advanced AI model).
- Qwen-3 Max was trained on a whopping 36 trillion tokens (these tokens are words that an AI reads while getting trained).
- The capabilities of this model are such that it can read, remember and process huge amounts of text all at once (let’s say, an entire book, long research paper, or even full codebases).
- In technical terms, it can take a context length of up to 1 million tokens.
What Can Qwen-3 Max Do Better Now?
According to Alibaba, the Qwen-3 Max is improved in:
- Reasoning (meaning figuring things out step by step).
- Instruction following (the model can understand and carry out all the tasks you give it).
- Multilingual support (for now, the model can work well in both English and Chinese).
- Long-tail knowledge coverage (meaning the AI now knows about more complex subjects that are less common).
- Math, coding, logic, and science (this version is now trained to solve any technical tasks more accurately).
- Reduced hallucinations (this version is less likely to make things up as compared to its older models).
- High-quality response to open-ended or complex questions.
How Does It Rank Compared to Others?
Lmarena’s Text Leaderboard
- Qwen-3 Max is ranked third on the LMArena’s text leaderboard (it’s a ranking given to AI text-based models.
- It is ahead of GPT-5.
- And a little behind Gemini 2.5 Pro, Claude Opus 4.1 (thinking mode), OpenAI’s o3, 4-o, GPT-4.5, and GPT-5 High models.
SWE-Bench Verified
- Qwen-3 Max has scored 69.6. on SWE-Bench Verified (it is a test for AI models to rank their real coding problem-solving skill).
- It has scored better than DeepSeek V3.1 and just a little below Claude Opus 4.
Tau2-Bench
- Qwen-3 Max has scored 74.8 on the Tau2-Bench (it’s a test for AI to see how well the uses tools in an agent-like way).
- In this test, Qwen-3 is well ahead of Claude Opus 4 and DeepSeek V3.1.
How Can You Use Qwen-3 Max?
- Qwen is an open-source AI model. The new Qwen-3 Max is also free.
- One can access the new version on the Qwen app (available on iOS and Android) or directly on the Qwen website.
- If you are a new user, the app will default to install as Qwen-3 Max.
- In case you don’t see the new version, here’s what you can do:
- Tap on the top left corner of the Qwen app and then select the Qwen-3 Max model, and there you go.
- It’s just like ChatGPT or Gemini, you can interact and ask the AI any question you have.