Teaching Language Models to Think in Code

TL;DR

ThinC introduces a code-centric reasoning framework for language models, replacing NL-based reasoning with code execution, leading to improved performance on math benchmarks.

Contribution

It proposes a novel approach where code acts as the primary reasoner, outperforming existing tool-integrated reasoning methods on multiple benchmarks.

Findings

ThinC-4B outperforms all TIR baselines on five math benchmarks.

99.2% of answers are grounded in interpreter output.

Model recovers reliably from code execution failures.

Abstract

Tool-integrated reasoning (TIR) has emerged as a dominant paradigm for mathematical problem solving in language models, combining natural language (NL) reasoning with code execution. However, this interleaved setup has three key limitations: code often acts as a post-hoc verifier, intermediate NL computations are error-prone, and NL and code play overlapping rather than clearly distinct roles. We propose ThinC (Thinking in Code), a framework in which code itself serves as the reasoner rather than as a tool invoked by NL. A ThinC trajectory begins with a brief NL planning step, after which all reasoning unfolds through code blocks connected only by their execution outputs. We distill 12.2k code-centric trajectories from a teacher model and train ThinC-1.7B and ThinC-4B with supervised fine-tuning followed by reinforcement learning. ThinC-4B consistently outperforms every TIR baseline on five competition-level math benchmarks and even surpasses the much larger Qwen3-235B-A22B-Thinking. Further analysis shows that ThinC reasons through code: 99.2% of its final answers are grounded in interpreter output, and the model recovers reliably from code execution failures without intermediate NL reasoning. Our code and models will be released soon.