Evaluation is All You Need: Strategic Overclaiming of LLM Reasoning Capabilities Through Evaluation Design
This highlights a critical reproducibility problem for researchers and practitioners relying on benchmark claims for model selection and comparison.
The study found that benchmark evaluation results for reasoning models like Deepseek-R1-Distill and QwQ-32B are highly sensitive to subtle differences in evaluation conditions, leading to significant fluctuations and making claimed performance improvements difficult to reproduce reliably.
Reasoning models represented by the Deepseek-R1-Distill series have been widely adopted by the open-source community due to their strong performance in mathematics, science, programming, and other domains. However, our study reveals that their benchmark evaluation results are subject to significant fluctuations caused by various factors. Subtle differences in evaluation conditions can lead to substantial variations in results. Similar phenomena are observed in other open-source inference models fine-tuned based on the Deepseek-R1-Distill series, as well as in the QwQ-32B model, making their claimed performance improvements difficult to reproduce reliably. Therefore, we advocate for the establishment of a more rigorous paradigm for model performance evaluation and present our empirical assessments of the Deepseek-R1-Distill series models.