How Does Alibaba’s Tongyi Qianwen Compare to ChatGPT? A Hands‑On Evaluation

Official Introduction

Alibaba announced its ChatGPT‑like model “Tongyi Qianwen” and released the first batch of internal test invitation codes, allowing early users to experience the model.

Self‑Introduction

The model was prompted to introduce itself, responding with a brief self‑description.

Code Generation

Several programming tasks were submitted, including implementing bubble sort in Python, repeating a 19‑letter sequence 106 times, generating Fibonacci numbers recursively, writing logistic‑regression training code, and implementing a C‑style dynamic vector. The generated code snippets (shown in the accompanying images) were largely correct and matched the requested language, demonstrating strong code‑generation capability, though the model’s reasoning for more complex logic sometimes needed improvement.

“Alibaba’s model can understand common code requests, generate appropriate code, distinguish language requirements, and handle both English and Chinese prompts. Simple sorting and domain‑specific logistic‑regression code are generated well, but deeper reasoning still has room for growth.” – CSDN‑AI team director

Literary Creation

When asked to continue the classic tortoise‑and‑hare story, the model provided a correct ending and suggested several angles for further development, showing decent creative writing ability.

Mathematical Logic

Classic problems such as the chicken‑rabbit puzzle, a mother‑son age problem, and a workforce redistribution equation were posed. The model produced correct answers with concise explanations for the simpler tasks, but made a minor formatting error in a more complex equation, indicating that while its arithmetic reasoning is solid, handling intricate algebraic expressions can be error‑prone.

Chinese Understanding

Questions about the idiom “洛阳纸贵” and the roles in Beijing opera (sheng, dan, jing, chou) received brief yet accurate explanations, confirming reliable Chinese language comprehension.

Casual Chatting

For informal prompts—such as recommending a low‑cost dish, discussing the impact of AI on programmers, and general small‑talk—the model responded helpfully, providing detailed recipe steps and a neutral perspective on programmer impact.

Conclusion

Overall, Tongyi Qianwen demonstrates strong performance in code generation, basic mathematical reasoning, and Chinese language understanding, comparable to other leading LLMs, while still having room for improvement in complex logical reasoning and deeper conversational nuance.