DeepSeek's family of large language models (LLMs) may have put the spotlight on China's AI ambitions, but Alibaba and its Qwen efforts may win out.
Alibaba's Qwen3 launched and its flagship model, Qwen3-235B-A22B, is competitive with DeepSeek-R1, OpenAI's o1, o3-mini, Grok 3 and Gemini 2.5 Pro.
Two mixture of expert models, Qwen3-235B-A22B and Qwen3-30B-A3B, were open weighted and there are six other models under Apache 2.0 licenses.
Qwen3 introduces a hybrid approach to problem solving and support a thinking mode, where the model takes time to reason step by step, and non-thinking mode for quick answers for simple questions.
- Google Gemini vs. OpenAI, DeepSeek vs. Qwen: What we're learning from model wars
- Alibaba's cloud unit garners Q3 AI demand boost, touts Qwen efforts
- OpenAI launches o3 reasoning model, o4-mini as it aims for 'more agentic ChatGPT'
What that hybrid approach means is that users control how much thinking the model has to use. The hybrid approach can also work well with budgets and compute resources. The models also support 119 languages and dialects. In addition, Qwen3 was trained on about 36 trillion tokens.
In the short term, Qwen3 is the latest development in the LLM race in an almost daily game of leapfrog. In the long run, Qwen3 has a few built in advantages. First, the LLMs are backed by Alibaba. Second, Qwen3 seems to be popular on Hugging Face and has some open source heft. And finally, Qwen is easily accessible on Alibaba Cloud, which is one of the dominant cloud providers to enterprises in Asia.
Add it up and Qwen has some Alibaba ground game advantages that may be more relevant to enterprises.