LLM-Generated Python Code Samples from Google Code Jam

This dataset contains 5,575 Python code samples generated by 9 (including 2 different snapshots for GPT-3.5) large language models (LLMs) in response to 710 Google Code Jam competition problems. It was created to support the study of temporal drift, cross-model generalization, and stylometric detection of LLM-generated code.

Code samples were manually collected from ChatGPT (gpt-3.5-turbo) during May–June 2023. Additional samples were collected during June–July 2025 via OpenRouter APIs for the following models: gpt-3.5-turbo-1106, gpt-4o-mini, gpt-5-codex, claude-sonnet-4, codestral-2501, gemini-2.5-flash-preview-05-20, llama-3.3-8b-instruct, and qwen-2.5-coder-32b-instruct.

The dataset is released to promote reproducible research on AI-generated code and software provenance. It supplements an accepted paper at the 2026 IEEE Conference on Computing, Communications and Cybersecurity (CCWC); the paper DOI will be added once available. Documentation, metadata, and integrity hashes are included to support reuse and verification.