Alibaba's Qwen 3 is Better than OpenAI's o1 and o3-mini

Chinese tech giant Alibaba has launched the Qwen3 family of open-weight AI models, featuring a flagship 235B parameter model and a range of smaller versions, from 0.6B to 32B parameters.

The 235B model, with 22B activated parameters, can outperform OpenAI’s o1 and o3-mini models and rivals Google’s Gemini 2.5 Pro in key benchmarks.

However, it trails behind OpenAI’s latest o4-mini (high) model in coding and math tasks.

Qwen3 models can be deployed locally via Ollama, LM Studio, or accessed online through Qwen Chat, offering flexible "thinking" and "non-thinking" modes.

The models feature hybrid thinking modes for flexible problem-solving. "Thinking Mode" enables step-by-step reasoning for complex tasks, ensuring deeper, more accurate answers.

"Non-Thinking Mode" delivers fast, instant responses for simpler queries where speed matters most. This dual approach tailors performance based on the complexity of the task.

Moreover, the model supports 119 languages and dialects, which opens up new possibilities for international applications, enabling users worldwide to benefit from the power of these models.

"We have optimized the Qwen3 models for coding and agentic capabilities, and also we have strengthened the support of MCP as well. Below we provide examples to show how Qwen3 thinks and interacts with the environment," the company said.