Research map / Large language models / Agents, code and math
InteractBench: Benchmarking LLMs on Competitive Programming under Unrevealed Information
Jiaze Li, Aocheng Shen, Bing Liu, Boyu Zhang, Xiaoxuan Fan, Qiankun Zhang, Xianjun Deng
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 |
|---|---|---|---|
| add explicit reasoning (Thinking mode) to DeepSeek-V3.2 (ablation) | DeepSeek-V3.2 · InteractBench overall | pass@5: 0.262 -> 0.513 | Improved |
| add reasoning to DeepSeek-V3.2 on Graph category (ablation) | DeepSeek-V3.2 · InteractBench Graph | pass@5: 0.261 -> 0.511 | Improved |
| add thinking mode to Qwen3-32B (vs NonThinking) (ablation) | Qwen3-32B · InteractBench overall | pass@1: 0.017 -> 0.096 | Improved |
| add thinking mode to Qwen3-14B (vs NonThinking) (ablation) | Qwen3-14B · InteractBench overall | pass@5: 0.037 -> 0.121 | Improved |
| add reasoning to small Qwen3-14B on Hard problems (ablation) | Qwen3-14B · InteractBench Hard | pass@5: 0.000 | No clear effect |
| resample 5 solutions instead of 1 on Hard | Gemini-3-Pro-Preview · InteractBench Hard | pass: 0.367 -> 0.597 | Improved |
| resample 5 solutions instead of 1 on Hard | GPT-5.2 · InteractBench Hard | pass: 0.340 -> 0.492 | Improved |
| move from Easy to Hard problems | Gemini-3-Pro-Preview · InteractBench difficulty tiers | pass@5: 0.961 -> 0.597 | Hurt |
| move from Bit to Graph category | Gemini-3-Pro-Preview · InteractBench categories | pass@1: 0.613 -> 0.475 | Hurt |
Closest papers by meaning
- LiveCodeBench Pro: How Do Olympiad Medalists Judge LLMs in Competitive Programming? (NeurIPS 2025 · similarity 0.91)
- ICPC-Eval: Probing the Frontiers of LLM Reasoning with Competitive Programming Contests (NeurIPS 2025 · similarity 0.90)
- ShortcutsBench: A Large-Scale Real-world Benchmark for API-based Agents (ICLR 2025 · similarity 0.89)
- CS-Bench: A Comprehensive Benchmark for Large Language Models towards Computer Science Mastery (ICLR 2025 · similarity 0.89)
- ConvCodeWorld: Benchmarking Conversational Code Generation in Reproducible Feedback Environments (ICLR 2025 · similarity 0.88)
- BigCodeBench: Benchmarking Code Generation with Diverse Function Calls and Complex Instructions (ICLR 2025 · similarity 0.88)
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.
