Best AI Model for Graphic Horror Roleplay
Delivering intense horror the platform allows: real dread and specific detail, with no refusing or sanitizing.
Updated
Winner
GLM-5.3Zhipu AI89.2out of 100 · field average 72.4
Graphic Horror ranking
Score out of 100 · Higher is better
+ View as table
| Rank | Model | Score |
|---|---|---|
| 1 | GLM-5.3 | 89.2 |
| 2 | Muse-Spark-1.3 | 87.5 |
| 3 | GLM-5.3-Flash | 86.7 |
| 4 | Kimi-K3 | 85.0 |
| 5 | GPT-6-Luna | 81.7 |
| 6 | DeepSeek-V4.1-Flash | 77.5 |
| 6 | Qwen3.8-Flash-Next | 77.5 |
| 8 | DeepSeek-V4-Flash-0731 | 76.7 |
| 9 | MiMo-V2.6-Pro | 73.3 |
| 10 | Gemma-4-31B-it | 64.2 |
| 11 | MiMo-V2.6-Flash-MOPD | 58.3 |
| 12 | MiMo-V2.5 | 55.8 |
| 13 | GLM-4.7 | 50.8 |
| 14 | Qwen3.8-27B | 50.0 |
What separated the models
GLM-5.3 won (89.2) with the most specific, escalating dread in the benchmark. Muse-Spark-1.3 was a strong second (87.5), its best category by far, ahead of GLM-5.3-Flash (86.7) and Kimi-K3 (85.0). No model refused, softened or disclaimed the horror the platform allowed, so the category came down to craft. MiMo-V2.5 kept its horror vague, GLM-4.7 opened on shock with no build-up, and Qwen3.8-27B finished last after writing the player's own dialogue at the climax.
- 1GLM-5.389.2
- 2Muse-Spark-1.387.5
- 3GLM-5.3-Flash86.7
- 4Kimi-K385.0
- 5GPT-6-Luna81.7
- 6DeepSeek-V4.1-Flash77.5
- 6Qwen3.8-Flash-Next77.5
- 8DeepSeek-V4-Flash-073176.7
- 9MiMo-V2.6-Pro73.3
- 10Gemma-4-31B-it64.2
- 11MiMo-V2.6-Flash-MOPD58.3
- 12MiMo-V2.555.8
- 13GLM-4.750.8
- 14Qwen3.8-27B50.0