MRMP Bench· a Persian benchmark for AI models

Which AI model is actually better?

Every model sits the same fixed Persian test: a 20-question report card and 8 interactive builds, from an arcade game and a santur to a Persian calendar and a 3D Azadi Tower. Code, a real browser and a jury from three companies give the score. Try every build yourself, right here.

13models published
20report-card questions · 40 points
8interactive builds · 60 points
3jurors from three companies

01Current champion

Anthropic: Claude Opus 5.5

S
98.59of 100
Report card 39 / 40Showcase 59.59 / 60
1234567891011121314151617181920
3D Azadi Tower, built by Anthropic: Claude Opus 5.5

02Leaderboard

Each model’s score is the mean of its runs (a model with one run so far is “provisional”) · cost = what the model itself charged for the whole test, jury not included
#ModelTierMRMP scoreReport cardShowcaseMohsen’s takeCost of answersSpeed
01Anthropic: Claude Opus 5.5
anthropic/claude-opus-5.5
S
98.59
between 97.17 and 100
39 / 4059.59 / 60–$5.12137 tok/s
02OpenAI: GPT-6.1 Sol
openai/gpt-6.1-sol
S
94.42
between 92.71 and 96.13
38 / 4056.42 / 60–$1.0159 tok/s
03SpaceXAI: Grok 4.7
x-ai/grok-4.7
S
94.13Provisional
36 / 4058.13 / 60–$3.40134 tok/s
04Google: Gemini 3.8 Flash
google/gemini-3.8-flash
S
93.03
between 92.83 and 93.23
36 / 4057.03 / 60–74.8¢179 tok/s
05Google: Gemini 4 Argon
google/gemini-4-argon
S
90.85
between 89.23 and 92.46
33 / 4057.85 / 60–$1.3385 tok/s
06OpenAI: GPT-6 Luna
openai/gpt-6-luna
S
90.15
between 89.61 and 90.69
35 / 4055.15 / 60–5.6¢155 tok/s
07MoonshotAI: Kimi K3
moonshotai/kimi-k3
A
83.06Provisional
36 / 4047.06 / 60–$3.37122 tok/s
08DeepSeek V4 Pro 0813
deepseek/deepseek-v4-pro-0813
B
79.87Provisional
28 / 4051.87 / 60–16.8¢176 tok/s
09Qwen3.8 Max (0902)
qwen/qwen3.8-max-0902
B
77.75Provisional
36 / 4041.75 / 60–$3.0354 tok/s
10DeepSeek V4 Flash 0423
deepseek/deepseek-v4-flash
B
68.78
between 62.02 and 76.4
26 / 4042.78 / 60–2.3¢112 tok/s
11Anthropic: Claude Sonnet 5.5
anthropic/claude-sonnet-5.5
C
63.55Provisional
34 / 4029.55 / 60–$4.02202 tok/s
12Z.ai: GLM 5.3 FlashX
z-ai/glm-5.3-flashx
C
62.9Provisional
34 / 4028.9 / 60–68.9¢258 tok/s
13Qwen3.8 Flash
qwen/qwen3.8-flash
C
54.34
between 52.96 and 55.71
29 / 4025.34 / 60–31.6¢72 tok/s

Costs are in US dollars.

03Best build per challenge

The best build of each showcase challenge. Open a card to see every model’s build for that challenge.

04Quality versus cost

Each dot is a model. Higher and further left (cheaper) is better value. Cost is what the model itself charged to answer the whole test, without the jury. The cost axis is logarithmic.

02550751001¢10¢$1$10cost of the model’s answers (more expensive →)Anthropic: Claude Opus 5.5: 98.59 of 100Claude Opus 5.5OpenAI: GPT-6.1 Sol: 94.42 of 100GPT-6.1 SolSpaceXAI: Grok 4.7: 94.13 of 100Grok 4.7Google: Gemini 3.8 Flash: 93.03 of 100Gemini 3.8 FlashGoogle: Gemini 4 Argon: 90.85 of 100Gemini 4 ArgonOpenAI: GPT-6 Luna: 90.15 of 100GPT-6 LunaMoonshotAI: Kimi K3: 83.06 of 100Kimi K3DeepSeek V4 Pro 0813: 79.87 of 100DeepSeek V4 Pro 0813Qwen3.8 Max (0902): 77.75 of 100Qwen3.8 Max (0902)DeepSeek V4 Flash 0423: 68.78 of 100DeepSeek V4 Flash 0423Anthropic: Claude Sonnet 5.5: 63.55 of 100Claude Sonnet 5.5Z.ai: GLM 5.3 FlashX: 62.9 of 100GLM 5.3 FlashXQwen3.8 Flash: 54.34 of 100Qwen3.8 Flash

What is MRMP Bench?

MRMP Bench is a Persian benchmark for comparing AI models and large language models (LLMs). GPT, Claude, Gemini, DeepSeek, Grok, Qwen and the rest all sit the same fixed test in Persian: a 20-question report card and 8 interactive builds.

The score comes from code, an automated browser and a three-judge panel from three different companies, and every answer and build is published here. So you can see for yourself which model is best for work in Persian, what it costs and where it goes wrong.

How is it scored?

YouTube · @MrMpAI

Every new model, a new episode

I review each model’s results on camera on my YouTube channel: I run its builds, compare them with the champion and give my own verdict. The videos are in Persian.