Has anyone been able to replicate OpenAI's results?

[David-Lee-1990]

https://openai.com/index/introducing-chatgpt-agent/#:~:text=On%20DSBench%E2%81%A0,a%20new%20window)

[xyang0]

Our continued experiments show that recent long CoT reasoning models perform significantly better on data analysis tasks, but still faces difficulties when directly applied to data modeling tasks.

Rather than model-only usage, we speculate OpenAI has a strong agent framework (scaffold) on data scientist tasks, and we encourage the community to continue to explore and improve their agent framework on these tasks. We have observed a few recent work pursuing along this line. For example:

DatawiseAgent: A Notebook-Centric LLM Agent Framework for Automated Data Science
https://arxiv.org/abs/2503.07044