EdgeBench is a real-world learning benchmark dataset for intelligent agents released by ByteDance Seed in 2026. It aims to evaluate the ability of autonomous AI agents to learn from real-world environments. Related research papers include... EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments . This dataset contains 134 real-world tasks, 51 of which are open source, covering six ability categories: scientific computing and machine learning, systems and software engineering, optimization, knowledge reasoning, formal reasoning, and game theory.

Data fields:

• task_id: A unique identifier for the task.
• name: The readable title of the task
• category: The category to which the task belongs
• Description: A detailed description of the task.
• language: The programming language required by the task.
• Metric: The method of scoring the task
• internet: A boolean value indicating whether internet access is allowed during task execution.

Citation

@misc{edgebench2026,
title  = {EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments},
author = {Deyao Zhu and Xin Zhou and Shengling Qin and Xuekai Zhu and Hangliang Ding and Shu Zhong and others},
year   = {2026},
url    = {https://arxiv.org/abs/2607.05155},
}