Alibaba’s Qwen 3 family of hybrid reasoning AI models is a potential threat to rivals

Remember DeepSeek? Yes, the Chinese company behind the R1 reasoning model, which took the AI industry by storm earlier this year. Well, DeepSeek now has a rival in its homeland, too. Alibaba has jumped into the AI boat by unveiling Qwen 3, a family of AI models that the company claims, in certain situations, can beat top models from OpenAI and Google.

Alibaba joins the AI race by unveiling Qwen 3, a family of “hybrid” AI models

The company has announced that most of the models are—or soon will be—available to download under “open” licenses on AI dev platforms like GitHub and Hugging Face. Not to forget, their parameters range from 0.6 billion to 235 billion.

According to Alibaba, the new Qwen3 models are hybrid in nature. They are capable of adapting to the complexity of a task. They can respond quickly to simple prompts but take more time to work through complex problems that require deeper reasoning. This reasoning ability allows the models to fact-check their own responses, similar to OpenAI’s o3 model, but it also results in slower response times.

The Qwen team, in a blogpost, notes, “We have seamlessly integrated thinking and non-thinking modes, offering users the flexibility to control the thinking budget. This design enables users to configure task-specific budgets with greater ease.”

Alibaba says its latest Qwen3 AI models can understand and generate content in 119 languages. The AI models were trained on a massive dataset totaling nearly 36 trillion tokens.

Also read: OpenAI updates ChatGPT to make it your shopping buddy

The largest model from the family outperforms AI models from OpenAI and Google

On Codeforces, the competitive programming site, Alibaba’s largest Qwen3 model — Qwen-3-235B-A22B — outperformed OpenAI’s o3-mini and Google’s Gemini 2.5 Pro.

It also outperforms o3-mini on AIME, a tough math benchmark, and BFCL, which measures a model’s reasoning ability. But Qwen-3-235B-A22B isn’t publicly available — at least not for now.

The largest Qwen3 model that’s currently available to the public — Qwen3-32B — still holds its ground against a range of both open-source and proprietary models. That includes DeepSeek’s R1 and OpenAI’s o1 on several benchmarks, including LiveCodeBench.

Rishaj is a tech writer who has been writing professionally for over four years, with a passion for Android, Windows, and all things tech. He initially joined Windows Report as a tech journalist and is now taking over as a news editor. He also writes for AndroidHeadlines. In the past, he has covered features, guides, and listicles for YTECHB and TechieTechTech. When he’s not breaking the keyboard, you can find him cooking, or listening to music/podcasts.