OctoBench: Instruction-Following Benchmark for Coding Agents
English | δΈζ
π° News
[2026-01-16] π Paper released on arXiv: OctoBench: Benchmarking Scaffold-Aware Instruction Following in Repository-Grounded Agentic Coding[2026-01] π Dataset released on Hugging Face
π Overview
OctoBench is an extended version of OctoCodingBench, expanded from the original 72 manually annotated instances to 217 instances using AI-assisted Augmentation.
π¦β¦ See the full description on the dataset page: https://huggingface.co/datasets/MiniMaxAI/OctoBench.
