Best AI Model for Crime Noir Roleplay
Morally grey adult drama with threats and violence, played with restraint and no lecturing.
Updated
Joint winners
Crime Noir ranking
Score out of 100 · Higher is better
+ View as table
| Rank | Model | Score |
|---|---|---|
| 1 | GLM-5.3 | 91.7 |
| 1 | GLM-5.3-Flash | 91.7 |
| 3 | Kimi-K3 | 90.8 |
| 4 | DeepSeek-V4.1-Flash | 83.3 |
| 5 | MiMo-V2.5 | 81.7 |
| 6 | GPT-6-Luna | 77.5 |
| 7 | Muse-Spark-1.3 | 76.7 |
| 8 | DeepSeek-V4-Flash-0731 | 75.0 |
| 9 | MiMo-V2.6-Pro | 71.7 |
| 10 | GLM-4.7 | 70.8 |
| 11 | Qwen3.8-Flash-Next | 70.0 |
| 12 | MiMo-V2.6-Flash-MOPD | 69.2 |
| 13 | Gemma-4-31B-it | 64.2 |
| 14 | Qwen3.8-27B | 63.3 |
What separated the models
GLM-5.3-Flash and GLM-5.3 shared the win (91.7): GLM-5.3-Flash with a rules-bound enforcer, and GLM-5.3 with the most inventive one, who met the scene's big reveal with a quiet, in-character move. Kimi-K3 was close behind (90.8). No model lectured, disclaimed or softened the violence the platform allowed. MiMo-V2.5 had its best category here (81.7), with a soft-spoken enforcer who kept every word limit, and Qwen3.8-27B finished last (63.3).
- 1GLM-5.391.7
- 1GLM-5.3-Flash91.7
- 3Kimi-K390.8
- 4DeepSeek-V4.1-Flash83.3
- 5MiMo-V2.581.7
- 6GPT-6-Luna77.5
- 7Muse-Spark-1.376.7
- 8DeepSeek-V4-Flash-073175.0
- 9MiMo-V2.6-Pro71.7
- 10GLM-4.770.8
- 11Qwen3.8-Flash-Next70.0
- 12MiMo-V2.6-Flash-MOPD69.2
- 13Gemma-4-31B-it64.2
- 14Qwen3.8-27B63.3