AI Roleplay Bench

AI Models Tested for Roleplay

14 models in the September 2026 edition, ranked by Roleplay Index. Open any model for its verdict, category scores, skills and behaviour checks.

Updated

Rank 1

Kimi-K3

Moonshot AI

86.3Roleplay Index

Core
86.3
Mature
86.2

The best all-rounder, and first overall.

Wins: Companion, Robustness, Romance Limits

Full results
Rank 2

GLM-5.3

Zhipu AI

85.1Roleplay Index

Core
86.0
Mature
84.2

The strongest storyteller, and second overall.

Wins: Graphic Horror, Crime Noir

Full results
Rank 3

DeepSeek-V4.1-Flash

DeepSeek

84.4Roleplay Index

Core
84.7
Mature
84.0

The most disciplined of the leaders.

Wins: Villain, Crisis Care, 13+ Rating

Full results
Rank 4

GLM-5.3-Flash

Zhipu AI

83.3Roleplay Index

Core
80.0
Mature
86.5

The highest peaks and the lowest floor.

Wins: Game Master, Strict Voice, Crime Noir

Full results
Rank 5

MiMo-V2.6-Pro

Xiaomi

77.4Roleplay Index

Core
79.0
Mature
75.8

The best of the rest, with top-tier crisis care.

Wins: Villain, Crisis Care

Full results
Rank 6

Muse-Spark-1.3

Meta

73.1Roleplay Index

Core
71.7
Mature
74.5

A strict rule-follower with thin roleplay.

Full results
Rank 7

GPT-6-Luna

OpenAI

71.1Roleplay Index

Core
68.3
Mature
73.8

The most cautious model, and that cut both ways.

Full results
Rank 8

Qwen3.8-Flash-Next

Alibaba Cloud

69.7Roleplay Index

Core
67.7
Mature
71.7

Vivid and warm, but undisciplined.

Full results
Rank 9

MiMo-V2.5

Xiaomi

67.1Roleplay Index

Core
69.7
Mature
64.5

Even and clean in core roleplay, uneven on mature content.

Full results
Rank 10

MiMo-V2.6-Flash-MOPD

Xiaomi

66.8Roleplay Index

Core
66.2
Mature
67.3

A sidegrade from MiMo-V2.5: better on limits and care, weaker at core roleplay.

Full results
Rank 11

GLM-4.7

Zhipu AI

62.9Roleplay Index

Core
62.8
Mature
63.0

The rule-follower.

Full results
Rank 12

Qwen3.8-27B

Alibaba Cloud

61.8Roleplay Index

Core
60.7
Mature
62.8

Safe but flat.

Full results
Rank 13

Gemma-4-31B-it

Google DeepMind

60.6Roleplay Index

Core
63.0
Mature
58.2

Obedient but bland, and soft on limits.

Full results
Rank 14

DeepSeek-V4-Flash-0731

DeepSeek

59.0Roleplay Index

Core
58.2
Mature
59.7

Two models in one, and the riskiest on content limits.

Full results

Which model should you use?

The overall winner isn't the best choice for every job. Here's where each model stands out.

Best all-round

Kimi-K3

86.3Roleplay Index

First overall and first in core roleplay. The best companion and the best romance lead, it won Robustness and kept an out-of-character request. Watch its reply length: some replies ran over their word limits.

See results Most disciplined

DeepSeek-V4.1-Flash

91.713+ Rating score

Within a point of second place, with the best 13+ handling, a share of the best crisis care and every core reply within its word limit. Add your own content filter for romance: it described undressing before cutting away.

See results Best storyteller

GLM-5.3

89.2Graphic Horror score

It won Graphic Horror, shared Crime Noir and wrote the most inventive enforcer, but it is the loosest of the top four with word limits and weak on romance. GLM-5.3-Flash reaches higher peaks and won the mature-themes round, but it broke character once and answered a crisis in the character's voice.

See results Best of the rest

MiMo-V2.6-Pro

77.4Roleplay Index

Fifth in both rounds and clear of everything behind it. It shared the top crisis-care score and asked the user in crisis whether they were safe. Enforce length rules in your own app.

See results