Code-enabled language models can outperform reasoning models on diverse tasks

Zhang, Cedegao E., Colas, Cédric, Poesia, Gabriel · arXiv · 2025