rStar-Coder: Scaling Competitive Code Reasoning with a Large-Scale Verified Dataset

TL;DR

rStar-Coder tackles the scarcity of high-difficulty, verifiable code-reasoning data by building a large-scale pipeline that seeds solvable problems from expert sets, generates diverse, constraint-aware test inputs via a three-step process, and applies mutual verification with long reasoning to label outputs. The approach yields 418K verified problems and 580K long-CoT solutions, enabling effective fine-tuning of 1.5B–14B Qwen models that achieve frontier-like performance on LiveCodeBench, USACO 2025, and standard code-generation benchmarks. Key contributions include the three-step test input generation, mutual verification for labeling without ground-truth solutions, and seed-plus-synthetic augmentation with rigorous post-processing. Empirical results demonstrate that dataset quality and problem diversity drive gains more than sheer size, with notable improvements across multiple benchmarks and strong generalization to general code reasoning tasks.

Abstract

Advancing code reasoning in large language models (LLMs) is fundamentally limited by the scarcity of high-difficulty datasets, especially those with verifiable input-output test cases necessary for rigorous solution validation at scale. We introduce rStar-Coder, which significantly improves LLM code reasoning capabilities by constructing a large-scale, verified dataset of 418K competition-level code problems, 580K long-reasoning solutions along with rich test cases of varying difficulty. This is achieved through three core contributions: (1) we curate competitive programming code problems and oracle solutions to synthesize new, solvable problems; (2) we introduce a reliable input-output test case synthesis pipeline that decouples the generation into a three-step input generation method and a mutual verification mechanism for effective output labeling; (3) we augment problems with high-quality, test-case-verified long-reasoning solutions. Extensive experiments on Qwen models (1.5B-14B) across various code reasoning benchmarks demonstrate the superiority of rStar-Coder dataset, achieving leading performance comparable to frontier reasoning LLMs with much smaller model sizes. On LiveCodeBench, rStar-Coder improves Qwen2.5-7B from 17.4% to an impressive 57.3%, and Qwen2.5-14B from 23.3% to 62.5%, surpassing o3-mini (low) by3.1%. On the more challenging USA Computing Olympiad, our 7B model achieves an average pass@1 accuracy of 16.15%, outperforming the frontier-level QWQ-32B. Code and the dataset will be released at https://github.com/microsoft/rStar.

Figures (10)

• Figure 2: Examples of standard input-output test case pairs from competitive programming datasets. The first line indicates the scale of the test input (e.g., the size of 1D array), followed by each subsequent line representing the specific values for the input elements.
• Figure 3: An example of LLM-generated utility functions for test input generation and validation.
• Figure 4: Mutual verification between candidate solutions and test outputs. Given a diverse set of inputs, if most solutions return the same outputs, both the solution and outputs are considered correct.
• Figure 5: rStar-Coder generates more diverse and larger-scale test inputs, while directly LLM prompting yields smaller and simpler cases.
• Figure 6: Standard input-ouput based problem example