Alibaba launches Qwen3, a family of “hybrid” AI reasoning models
China's Alibaba has introduced Qwen3, a family of artificial intelligence models. Their size varies from 0.6 billion to 235 billion parameters.
China's Alibaba has introduced Qwen3, a family of artificial intelligence models. Their size varies from 0.6 billion to 235 billion parameters.
China's Alibaba has introduced Qwen3, a family of artificial intelligence models. Their size varies from 0.6 billion to 235 billion parameters.
Most of the models are available — or will soon be available — for download under an “open” license on AI developer platform Hugging Face and GitHub , TechCrunch reports .
According to Alibaba, Qwen3 are “hybrid” models — they can spend time “thinking” about complex problems or quickly respond to simpler queries.
The company has integrated “thinking” and “not thinking” modes, which will allow users to set budgets for specific tasks.
Some models also use a combination of expert architectures (MoE), which can be more computationally efficient for answering queries. MoE breaks tasks down into subtasks and delegates them to smaller, specialized “expert” models.
Qwen3 models support 119 languages and were trained on a dataset of over 36 trillion tokens. The company said Qwen3 was trained on a combination of textbooks, “question-answer pairs,” code snippets, AI-generated data, and more.
These improvements, along with others, significantly expand Qwen3’s capabilities compared to its predecessor, Qwen2 , Alibaba said. None of the Qwen3 models appear to outperform newer models such as OpenAI’s o3 and o4-mini , but they are nonetheless strong contenders.

On Codeforces, a platform for programming competitions, the largest Qwen3 model—Qwen-3-235B-A22B—beat OpenAI’s o3-mini and Google’s Gemini 2.5 Pro. Qwen-3-235B-A22B also outperformed o3-mini on the latest version of AIME, a complex math test, and BFCL, a test to assess a model’s ability to “reason” about problems. But Qwen-3-235B-A22B is not publicly available—at least not yet.
The second Qwen3 model, Qwen3-32B, which is public, outperforms OpenAI's o1 model in several tests, including the LiveCodeBench coding benchmark.
In addition to downloadable models, Qwen3 is available from cloud providers including Fireworks AI and Hyperbolic.


