---
title: "Kimi K3 on Role-Play Bench — SociaBench"
description: "Platform evaluation of Kimi K3 on Role-Play Bench: Overall 71.41 (out of 100), with scores by dimension, sample counts and evaluation times."
canonical: "https://sociabench.com/models/minimax-rp/moonshotai-kimi-k3"
markdown: "https://sociabench.com/models/minimax-rp/moonshotai-kimi-k3.md"
type: "WebPage / Dataset"
entity: "https://sociabench.com/models/minimax-rp/moonshotai-kimi-k3#evaluation"
about: "https://sociabench.com/b/minimax-rp#benchmark"
part_of: "https://sociabench.com/b/minimax-rp#leaderboard"
publisher: "SociaBench"
language: "en-US"
date_modified: "2026-11-01T14:44:06.466Z"
data_version: "all:22-1793544246466-r1793034246936-p2710 zh:22-1793544246466-r1793034246936-p2710 en:22-1793544246466-r1793034246936-p2710"
result_source: "Platform evaluation (Platform reproduction result)"
license: "https://sociabench.com/data-terms"
sources:
  - "Original source: MiniMax, Released January 2026 with the MiniMax-M2-her blog post, \"A Deep Dive into the MiniMax-M2-her\": https://www.minimax.io/news/a-deep-dive-into-the-minimax-m2-her-2"
  - "Test set and official implementation: https://huggingface.co/datasets/MiniMaxAI/role-play-bench"
  - "Reproduction report (SociaBench): https://sociabench.com/b/minimax-rp/methodology"
---

# Kimi K3 · Role-Play Bench

Results of Kimi K3 on [Role-Play Bench](https://sociabench.com/b/minimax-rp). Results SociaBench obtained on this platform by running the full test set under the benchmark's current reproduction protocol. Only these results are ranked.

## Results

| View | Rank | Overall | 95% CI | Samples | Model version | Evaluated at | Data version |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Overall | 11 / 11 | 71.41 | [70.46, 72.39] | 285 | — | 2026-11-01T14:44Z | 22-1793544246466-r1793034246936-p2710 |
| Chinese | 10 / 11 | 70.31 | [69.12, 71.50] | 150 | — | 2026-11-01T14:44Z | 22-1793544246466-r1793034246936-p2710 |
| English | 11 / 11 | 72.64 | [71.32, 74.02] | 135 | — | 2026-11-01T14:44Z | 22-1793544246466-r1793034246936-p2710 |

## Scores by group

| View | Worlds | Stories | Preferences |
| --- | --- | --- | --- |
| Overall | 62.31 | 68.24 | 92.80 |
| Chinese | 61.21 | 67.51 | 91.32 |
| English | 63.53 | 69.06 | 94.44 |

## Scores by dimension

| View | Basics | Logic | Knowledge | Diversity | Content Logic | Interaction |
| --- | --- | --- | --- | --- | --- | --- |
| Overall | 40.67 | 46.91 | 99.35 | 59.25 | 86.22 | 92.80 |
| Chinese | 38.07 | 45.87 | 99.69 | 57.63 | 87.28 | 91.32 |
| English | 43.56 | 48.07 | 98.97 | 61.06 | 85.04 | 94.44 |

## Sources and methods

- Original source: MiniMax, Released January 2026 with the MiniMax-M2-her blog post, "A Deep Dive into the MiniMax-M2-her": https://www.minimax.io/news/a-deep-dive-into-the-minimax-m2-her-2
- Test set and official implementation: https://huggingface.co/datasets/MiniMaxAI/role-play-bench
- Reproduction report (SociaBench): https://sociabench.com/b/minimax-rp/methodology
- Leaderboard: https://sociabench.com/b/minimax-rp

## Citation

If you repost or cite any evaluation result from this site, credit "SociaBench (AI Social Intelligence Leaderboards)" and link to this site: https://sociabench.com/models/minimax-rp/moonshotai-kimi-k3.