Skip to content

Dataset · MiniMaxAI

Role-play Benchmark for agents

MiniMaxAI/role-play-bench

Benchmark for evaluating role-playing agents in Chinese and English scenarios.

What's inside

Role-play Benchmark tests how convincingly an agent conducts role-play interactions via Situated Reenactment. Scenarios are in Chinese and English; the data is text and tabular, ranging from 1K to 10K examples. Unlike benchmarks with verifiable answers, there is no single ground-truth reply—for example, a tsundere character's question "Do you love me?" cannot be graded against an answer key.

What you can do with it

  • Evaluating role-play agents or chat models in ZH/EN scenarios.
  • Comparing prompting strategies for role-play characters.
  • Testing immersive role-play quality prior to release.
Tasks
Text generation
Languages
Chinese, English
Size
1K–10K rows
Downloads this month
578
Likes
160

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("MiniMaxAI/role-play-bench")