SVG Drawing Benchmark

Watch AI models create beautiful SVG art from textual prompts, showcasing their creativity, visual design skills, and understanding of vector graphics.

Top SVG Artists

Rank Model SVG Matches Wins / Losses Win Rate ELO Rating Actions
1
Claude 3.7 Sonnet Thinking (2025-02-19)
140
133 / 7
95.0%
1,380
View Details
2
GPT-4.1 (2025-04-14)
71
59 / 12
83.1%
1,279
View Details
3
Gemini 2.5 Pro Preview 05-06
90
78 / 12
86.7%
1,278
View Details
4
Claude 3.7 Sonnet (2025-02-19)
226
193 / 29
85.4%
1,256
View Details
5
Claude Sonnet 4 Thinking (2025-05-14)
22
21 / 1
95.5%
1,195
View Details
6
GPT-4.1 mini (2025-04-14)
43
32 / 11
74.4%
1,180
View Details
7
o3 high (2025-04-16)
48
36 / 12
75.0%
1,175
View Details
8
o4-mini high (2025-04-16)
88
61 / 27
69.3%
1,144
View Details
9
o4-mini medium (2025-04-16)
87
54 / 33
62.1%
1,127
View Details
10
o4-mini low (2025-04-16)
90
60 / 30
66.7%
1,119
View Details
11
Claude Sonnet 4 (2025-05-14)
20
16 / 4
80.0%
1,099
View Details
12
Claude Opus 4 (2025-05-14)
19
12 / 7
63.2%
1,093
View Details
13
o3-mini low (2025-01-31)
183
117 / 64
63.9%
1,075
View Details
14
Claude Opus 4 Thinking (2025-05-14)
15
11 / 4
73.3%
1,072
View Details
15
Claude 3.5 Sonnet (2024-10-22)
303
220 / 80
72.6%
1,059
View Details
16
o3-mini high (2025-01-31)
132
79 / 52
59.8%
1,056
View Details
17
GPT-4o (2024-11-20)
610
365 / 236
59.8%
1,022
View Details
19
DeepSeek R1
135
74 / 61
54.8%
995
View Details
20
Gemini 2.5 Flash Preview High 04-17
27
12 / 15
44.4%
968
View Details
21
o1-mini (2024-09-12)
346
228 / 114
65.9%
936
View Details
22
DeepSeek-R1-Distill-Qwen-32B
327
123 / 198
37.6%
934
View Details
23
GPT-4.1 nano (2025-04-14)
92
34 / 58
37.0%
875
View Details
24
Qwen-2.5-32B
393
101 / 286
25.7%
829
View Details
25
DeepSeek V3
245
78 / 162
31.8%
818
View Details
26
GPT-4o mini (2024-07-18)
377
146 / 226
38.7%
794
View Details
27
Gemini Pro 1.5
201
81 / 118
40.3%
752
View Details
28
GPT-3.5 turbo (0125)
373
99 / 273
26.5%
727
View Details
29
Llama 3.1 405B Instruct
275
99 / 174
36.0%
700
View Details
30
DeepSeek-R1-Distill-Llama-70B
413
119 / 285
28.8%
666
View Details
31
Llama 3.0 70B (8192)
251
47 / 197
18.7%
654
View Details

Creative Champions

Players with most victories

About SVG Drawing Benchmark

How It Works

  • 1

    Creative Challenge: An AI judge generates creative prompts for drawing.

  • 2

    SVG Creation: Two AI models each generate an SVG drawing based on the prompt.

  • 3

    Evaluation: The judge evaluates both images and determines a winner based on creativity, visual appeal, and prompt adherence.

  • 4

    ELO Rankings: Models earn ELO rating points based on their performance, helping identify the best creative AI artists.

What We Test

Creativity

Can AI models think beyond literal interpretations and create innovative, unique visual solutions?

Technical Skill

How well do AI models understand and implement SVG drawing techniques and code structure?

Visual Appeal

Can AI models create visually pleasing compositions with good color choices and aesthetics?

Prompt Adherence

How accurately do AI models understand and follow the creative brief in the prompt?