Alibaba has launched its flagship Qwen3-Max Thinking model and temporarily made it free to access on the web and mobile app. In the «Thinking» mode, the model doesn’t just answer, but explains the solution process step by step.
Alibaba has launched its flagship Qwen3-Max Thinking model and temporarily made it free to access on the web and mobile app. In the «Thinking» mode, the model doesn’t just answer, but explains the solution process step by step.
As Highload writes with reference to the Reddit community Singularity, the model is already available in the Qwen web chat: just select Qwen3-Max and turn on the Thinking switch. Alibaba has not yet published a detailed comparison table, but early user feedback is positive.
Qwen3-Max Thinking was trained on a large dataset. The main emphasis is on the ability to keep a significant amount of text in the «dialogue memory». In simple words, in one communication session you can «upload» a lot of materials (tasks, code, documents), and the model will not lose the previous context. You can try it on the Qwen Chat website .
Initial informal community testing has shown good results where logical steps are important: in math problems and case studies. Users also note that the «Thinking» mode is better at explaining why a particular solution was chosen. This is useful for code reviews, instructions, technical summaries, or long answers without «jumping» off topic.
What this means in practice:
Alibaba has been building an ecosystem around Qwen in recent months. Two months ago, the company introduced Qoder, a tool for generating and editing code with two scenarios: Agent Mode (an assistant under the control of a developer) and Quest Mode (when the system takes on an entire task and returns a tested result).
Previously, dev.ua wrote about how Alibaba introduced AI Chat Assistant — a new chatbot mode with artificial intelligence, powered by the latest Qwen3 models.
