Research map / Large language models / Agents, code and math
RECODE: A Benchmark for Research Code DEvelopment with Interactive Human Feedback
Chunyu Miao, Henry Peng Zou, Yangning Li, Yankai Chen, Yibo Wang, Fangxin Wang, Yifan Li, Wooseong Yang, Bowei He, Xinni Zhang, Dianzhi Yu, Hanchen Yang and 17 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| raise feedback from Level 0 to Level 1 (ablation) | GPT-5 · RECODE-H | Recall: 0.294 -> 0.451 | Improved |
| raise feedback from Level 0 to Level 4 (ablation) | DeepSeek-V3.1 · RECODE-H | Recall: 0.108 -> 0.706 | Improved |
| raise feedback from Level 0 to Level 4 (ablation) | GPT-5 · RECODE-H | Recall: 0.294 -> 0.716 | Improved |
| raise feedback from Level 2 to Level 3 (ablation) | Claude-sonnet-4 · RECODE-H | Recall: 0.333 -> 0.333 | No clear effect |
| raise feedback from Level 2 to Level 3 (ablation) | DeepSeek-V3.1 · RECODE-H | test case pass rate: 0.606 -> 0.576 | Hurt |
| raise feedback from Level 3 to Level 4 (add correct code snippet) (ablation) | GPT-5 · RECODE-H | test case pass rate: 0.837 -> 0.843 | Improved |
| raise feedback from Level 0 to Level 1 (ablation) | GPT-5-mini · RECODE-H | MRR: 0.127 -> 0.062 | Hurt |
| use GPT-5 instead of GPT-o3 as feedback model (ablation) | GPT-5-mini · RECODE-H | test case pass rate: 0.27 -> 0.32 | Improved |
| use a larger GPT-5 model instead of GPT-5-nano | GPT-5-nano · RECODE-H Level 4 | Recall: 0.353 -> 0.716 (nano vs GPT-5) | Improved |
| use Gemini-2.5-pro instead of Gemini-2.5-flash | Gemini-2.5 · RECODE-H | Recall: L2: 0.343 vs 0.373 | Mixed |
| richer feedback increases adoption | Claude-sonnet-4 · RECODE-H | adoption rate | Hurt |
Closest papers by meaning
- ConvCodeWorld: Benchmarking Conversational Code Generation in Reproducible Feedback Environments (ICLR 2025 · similarity 0.90)
- SciCode: A Research Coding Benchmark Curated by Scientists (NeurIPS 2024 · similarity 0.88)
- R2E: Turning any Github Repository into a Programming Agent Environment (ICML 2024 · similarity 0.87)
- MINT: Evaluating LLMs in Multi-turn Interaction with Tools and Language Feedback (ICLR 2024 · similarity 0.87)
- From Reproduction to Replication: Evaluating Research Agents with Progressive Code Masking (ICLR 2026 · similarity 0.87)
- NL2Repo-Bench: Towards Long-Horizon Repository Generation Evaluation of Coding Agents (ICML 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
