CRUXEval: A Benchmark for Code Reasoning, Understanding and Execution

TL;DR

CRUXEval is a new benchmark with 800 Python functions designed to evaluate code reasoning, understanding, and execution, revealing gaps in current models' capabilities and providing insights for future improvements.

Contribution

It introduces a generic recipe for creating execution benchmarks and evaluates twenty models, highlighting their limitations and the potential of simple schemes like CoT and fine-tuning.

Findings

Recent high-scoring models on HumanEval perform poorly on CRUXEval.

GPT-4 with CoT achieves up to 81% pass@1 on output prediction.

Open-source models like Code Llama 34B lag behind GPT-4 in performance.

Abstract

We present CRUXEval (Code Reasoning, Understanding, and eXecution Evaluation), a benchmark consisting of 800 Python functions (3-13 lines). Each function comes with an input-output pair, leading to two natural tasks: input prediction and output prediction. First, we propose a generic recipe for generating our execution benchmark which can be used to create future variation of the benchmark. Second, we evaluate twenty code models on our benchmark and discover that many recent high-scoring models on HumanEval do not show the same improvements on our benchmark. Third, we show that simple CoT and fine-tuning schemes can improve performance on our benchmark but remain far from solving it. The best setup, GPT-4 with chain of thought (CoT), achieves a pass@1 of 75% and 81% on input and output prediction, respectively. In contrast, Code Llama 34B achieves a pass@1 of 50% and 46% on input and output prediction, highlighting the gap between open and closed source models. As no model is close to acing CRUXEval, we provide examples of consistent GPT-4 failures on simple programs as a lens into its code reasoning capabilities and areas for improvement.