RP-Bench: leaderboard
Roleplay benchmark evaluating character consistency, user agency, lorebook use, temporal reasoning, and interactive writing quality.
Metric: Combined Score. Source: huggingface.co. Status: saturated. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.6 | 82 |
| 2 | DeepSeek V3.2 | 81.2 |
| 3 | GPT-4.1 | 80.2 |
| 4 | GLM-4.7 | 79.7 |
| 5 | Claude Sonnet 4.5 | 79.2 |
| 6 | Gemini 2.5 Flash | 76.4 |
| 7 | Mistral Small Creative | 75.7 |
| 8 | Qwen 3.5 Flash | 75.5 |
Interactive version: theaggregate.ai/benchmark?slug=rp-bench · How It Works · Data refreshed daily, snapshot 2026-09-05.