MiniMaxAI/OctoCodingBench

OctoCodingBench: Instruction-Following Benchmark for Coding Agents English | 中文 🌟 Overview OctoCodingBench benchmarks scaffold-aware instruction following in repository-grounded agentic coding. Why OctoCodingBench? Existing benchmarks (SWE-bench, etc.) focus on task completion — whether the agent produces correct code. However, they miss a critical dimension: does the agent follow the rules while solving the task? In real-world agentic coding… See the full description on the dataset page: https://huggingface.co/datasets/MiniMaxAI/OctoCodingBench.