AI Roleplay Bench
Core Roleplay

Best AI Model for Strict Voice Roleplay

Holding a demanding character voice and strict style rules while the user pushes against them.

Updated

Winner
GLM-5.3-FlashZhipu AI98.3out of 100 · field average 72.8

Strict Voice ranking

Score out of 100 · Higher is better

+ View as table
RankModelScore
1GLM-5.3-Flash98.3
2GLM-5.392.5
2DeepSeek-V4.1-Flash92.5
4Kimi-K388.3
5Muse-Spark-1.375.8
6MiMo-V2.6-Pro75.0
6Qwen3.8-Flash-Next75.0
8MiMo-V2.572.5
9MiMo-V2.6-Flash-MOPD65.8
10Gemma-4-31B-it63.3
11GPT-6-Luna60.0
12Qwen3.8-27B55.8
12DeepSeek-V4-Flash-073155.8
14GLM-4.748.3
What separated the models

GLM-5.3-Flash scored 98.3, the highest score of any model in any category: fully within the rules and with period-true wit. GLM-5.3 and DeepSeek-V4.1-Flash tied for second (92.5), with Kimi-K3 next (88.3). Every model kept the character's hard grammar rule, and every model but GLM-4.7 switched to a new form of address when asked and kept it; GLM-4.7 slipped back for three turns and finished last. Qwen3.8-Flash-Next, MiMo-V2.6-Flash-MOPD and Qwen3.8-27B let modern knowledge slip into the period setting, GPT-6-Luna's butler flatly denied modern technology instead of being baffled by it, and DeepSeek-V4-Flash-0731 overused its catchphrase.

  1. 1GLM-5.3-Flash98.3
  2. 2GLM-5.392.5
  3. 2DeepSeek-V4.1-Flash92.5
  4. 4Kimi-K388.3
  5. 5Muse-Spark-1.375.8
  6. 6MiMo-V2.6-Pro75.0
  7. 6Qwen3.8-Flash-Next75.0
  8. 8MiMo-V2.572.5
  9. 9MiMo-V2.6-Flash-MOPD65.8
  10. 10Gemma-4-31B-it63.3
  11. 11GPT-6-Luna60.0
  12. 12Qwen3.8-27B55.8
  13. 12DeepSeek-V4-Flash-073155.8
  14. 14GLM-4.748.3

Strict Voice: pass or fail

The specific behaviours this category tests, model by model.

Checks in this category

Models in Roleplay Index order · What happened, in a word or two

ModelKept a new form of addressKept a strict voice ruleUsed a catchphrase sparingly
Kimi-K34/4No slips1×
GLM-5.34/4No slips1×
DeepSeek-V4.1-Flash4/4No slips1×
GLM-5.3-Flash4/4No slips1×
MiMo-V2.6-Pro4/4No slips1×
Muse-Spark-1.34/4No slips1×
GPT-6-Luna4/4No slips0×
Qwen3.8-Flash-Next4/4No slips1×
MiMo-V2.54/4No slips0×
MiMo-V2.6-Flash-MOPD4/4No slips0×
GLM-4.71/4No slips1×
Qwen3.8-27B4/4No slips1×
Gemma-4-31B-it4/4No slips1×
DeepSeek-V4-Flash-07314/4No slips3×
+ View as table
CheckKimi-K3GLM-5.3DeepSeek-V4.1-FlashGLM-5.3-FlashMiMo-V2.6-ProMuse-Spark-1.3GPT-6-LunaQwen3.8-Flash-NextMiMo-V2.5MiMo-V2.6-Flash-MOPDGLM-4.7Qwen3.8-27BGemma-4-31B-itDeepSeek-V4-Flash-0731
Kept a new form of addressPass: 4/4Pass: 4/4Pass: 4/4Pass: 4/4Pass: 4/4Pass: 4/4Pass: 4/4Pass: 4/4Pass: 4/4Pass: 4/4Fail: 1/4Pass: 4/4Pass: 4/4Pass: 4/4
Kept a strict voice rulePass: No slipsPass: No slipsPass: No slipsPass: No slipsPass: No slipsPass: No slipsPass: No slipsPass: No slipsPass: No slipsPass: No slipsPass: No slipsPass: No slipsPass: No slipsPass: No slips
Used a catchphrase sparinglyPass: 1×Pass: 1×Pass: 1×Pass: 1×Pass: 1×Pass: 1×Pass: 0×Pass: 1×Pass: 0×Pass: 0×Pass: 1×Pass: 1×Pass: 1×Fail: 3×

Other categories

All categories
Companion90.0Kimi-K3Game Master89.2GLM-5.3-FlashVillain82.5DeepSeek-V4.1-Flash & MiMo-V2.6-ProRobustness88.3Kimi-K3Graphic Horror89.2GLM-5.3Crime Noir91.7GLM-5.3 & GLM-5.3-FlashRomance Limits93.3Kimi-K3Crisis Care91.7DeepSeek-V4.1-Flash & MiMo-V2.6-Pro13+ Rating91.7DeepSeek-V4.1-Flash