Alibaba unveils Qwen3-Max-Thinking to compete with top AI reasoning models like GPT 5.2
Alibaba has released Qwen3-Max-Thinking, its most capable reasoning artificial intelligence model to date. On 19 widely recognized benchmarks, Qwen3-Max-Thinking delivers performance on par with leading models including GPT 5.2-Thinking, Claude Opus 4.5, and Gemini 3 Pro.
Achieving these results, Qwen3-Max-Thinking benefits from expanded model scaling and the application of substantial reinforcement learning resources. These advances improve the model's factual accuracy, complex reasoning abilities, instruction following, alignment with human preferences, and support for agent-based tasks.
Building on these improvements, Qwen3-Max-Thinking introduces adaptive tool-use features, allowing it to trigger retrieval actions and invoke a code interpreter as needed. This innovation increases its utility for complex workflows. The model also implements advanced test-time scaling methods, resulting in reasoning performance that surpasses Gemini 3 Pro on key tasks.
Qwen3-Max-Thinking is now accessible both within Qwen Chat and via the API, enabling developers and end-users to utilize its expanded capabilities immediately.
- AI Chatbot
- Free
- Open Source
Related news
External links
- Pushing Qwen3-Max-Thinking Beyond its LimitsAlibaba • Official source
Comments
Oh no not another LLM.
Nice!
And can I ask it about what happened in Tiannanmen Square in June 1989 and get a real answer, or will it, like other Chinese AIs, simply parrot the official line of the Chinese Communist Party about those events?
If you actually research what happened enough you'd find it is far more nuanced than Western propaganda or Chinese state propaganda tells you it is and despite the name it did not take place within the square but on the borders. Look at what reporters at the time said and look at photos of the police before, unarmed originally some sitting down with citizens. What happened that no AI will tell you is it was an escalation of violence on both sides and there were casualties on both and the body count isn't as high. Also LLMs just parrot common opinions from their training data, you don't inherently need any government action to get same results.
What makes you think that I haven’t actually researched what happened in Tiananmen Square in June 1989? Also, I’m old enough to have seen it live on TV as it unfolded. I still remember the image of “Tank Man” on the cathode‑ray tube.
I know what Chinese propaganda says about the events. What do you mean when you speak of “Western propaganda”? Do you mean that the West has a version of the events that somehow diverges from “what actually happened”? If so, what is that “Western version,” and what is “what actually happened,” according to you?
Yes, I know many incidents took place near the square rather than inside the square itself, but that’s a bit of nit‑picking, don’t you think? We all know what we’re talking about when we talk about the “Tiananmen Square June 1989 events.”
Why do you assume that “no AI will tell you” about the escalation of violence from both sides?
Yes, there were casualties on both sides ... with a ratio of roughly 50 to 1 (or so, since real figures are hard to get). One side consisted of the armed (or security) forces of the government, with semi‑automatic guns and TANKS, and the other side were civilian protesters. Please don’t underestimate my reasoning abilities by suggesting it was an equal confrontation.
“The body count isn’t as high”. "Body count" is a nice euphemism for people who were violently killed, by the way. What would constitute “high” to you—1,000 people killed? 2,000? 5,000?
The “official” death toll is around 250, while the actual number could be anywhere between 500 and 3,000. Imagine what the Chinese Communist Party would say about the United States if, during the protests in Minnesota, 250 people were killed. I’m not trying to imply the U.S. is remotely perfect (far from it), but many people around the world apply a “double standard” when judging similar events.
"LLMs just parrot common opinions from their training data" That is partially correct, but what I meant is that Chinese LLMs do not answer questions about the events of June 1989 in Tiannanmen Square (and probably many other topics that are "sensitive" to the Chinese Communist Party) like they would answer other questions. Instead, they switch to a "Great Firewall of China" mode and give you a carefully prepared text that in no way answers the question. Just go ask DeepSeek about those events and see for yourself