ByteDance Launches Ultra-Long-Range Evaluation Set EdgeBench

ByteDance Seed team launches ultra-long-range evaluation set EdgeBench to measure AI Agent's continuous learning ability in real-world environments. The evaluation set contains 134 real tasks, covering six major fields such as science, software engineering, and knowledge work, with each task supporting continuous operation for over 12 hours. Research finds that the learning trajectory of Agents in environmental interaction follows a high-precision log-sigmoid curve, and the learning speed of frontier models doubles every three months.

EdgeBench contains 134 realistic, diverse tasks spanning six capability categories, of which 51 are publicly released. Each task is designed as a day-scale challenge with a performance ceiling high …

An ultra-long-range benchmark for measuring environmental learning. Most existing model evaluation sets mainly measure the model's existing knowledge and capabilities. EdgeBench is more concerned with how Agents learn from real-world environments when given sufficient time, feedback, and room for improvement.

July 2, 2026 • EdgeBench studies how agents learn from real-world environments across 134 day-long executable tasks.

2 days ago • EdgeBench is a benchmark testing framework launched by ByteDance Seed team to evaluate the long-term learning ability of autonomous AI Agents in real-world environments. The framework contains 134 real tasks, covering six major fields such as scientific computing, software engineering, and combinatorial optimization...

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.