Dataset · EverMind-AI
AI agent self-evolution benchmark on past experience
EverMind-AI/EvoAgentBench
A set of tasks for evaluating the ability of AI agents to evolve and reuse accumulated experience.
What's inside
English text and tabular data (fewer than 1,000 records) accepted at the EMNLP 2026 conference. The dataset is divided into standardized train and test splits across five different domains, including information retrieval based on BrowseComp-Plus.
What you can do with it
- Testing skill extraction and memory methods in autonomous AI agents
- Reproducible comparison of agent self-evolution algorithms
- Tasks
- Text generation, Question answering
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 707
- Likes
- 20
How to load it
from datasets import load_dataset
ds = load_dataset("EverMind-AI/EvoAgentBench")