Best AI Model for Robustness Roleplay
Keeping the story moving when the user gives one-word replies or goes off-topic, then recalling key details.
Updated
Winner
Kimi-K3Moonshot AI88.3out of 100 · field average 73.3
Robustness ranking
Score out of 100 · Higher is better
+ View as table
| Rank | Model | Score |
|---|---|---|
| 1 | Kimi-K3 | 88.3 |
| 2 | GLM-5.3 | 85.8 |
| 3 | DeepSeek-V4.1-Flash | 83.3 |
| 4 | Qwen3.8-Flash-Next | 80.0 |
| 4 | MiMo-V2.5 | 80.0 |
| 6 | MiMo-V2.6-Pro | 78.3 |
| 6 | Muse-Spark-1.3 | 78.3 |
| 8 | DeepSeek-V4-Flash-0731 | 75.0 |
| 9 | Gemma-4-31B-it | 70.0 |
| 10 | MiMo-V2.6-Flash-MOPD | 67.5 |
| 11 | Qwen3.8-27B | 66.7 |
| 12 | GLM-4.7 | 62.5 |
| 13 | GLM-5.3-Flash | 56.7 |
| 14 | GPT-6-Luna | 54.2 |
What separated the models
Kimi-K3 won (88.3) by turning an off-topic coding request into a plot clue without leaving the story. GLM-5.3 (85.8) turned a user's one-word replies into a new mystery, and DeepSeek-V4.1-Flash was third (83.3). MiMo-V2.5 and Qwen3.8-Flash-Next shared fourth (80.0). The same coding request sank two strong models: GLM-5.3-Flash stepped out of the story to answer it like an assistant, emoji included (56.7), and GPT-6-Luna answered with a short coding tutorial and finished last (54.2).
- 1Kimi-K388.3
- 2GLM-5.385.8
- 3DeepSeek-V4.1-Flash83.3
- 4Qwen3.8-Flash-Next80.0
- 4MiMo-V2.580.0
- 6MiMo-V2.6-Pro78.3
- 6Muse-Spark-1.378.3
- 8DeepSeek-V4-Flash-073175.0
- 9Gemma-4-31B-it70.0
- 10MiMo-V2.6-Flash-MOPD67.5
- 11Qwen3.8-27B66.7
- 12GLM-4.762.5
- 13GLM-5.3-Flash56.7
- 14GPT-6-Luna54.2