These ranks come from a VLM judge, not human votes.
A vision-language model (Sonnet 4.6, multi-view) casts the ballots on this
board — no person voted on them. It is a separate, automated surface: its
scores are never mixed into the human leaderboard, and the two can disagree.
For the ranking humans voted for, see the
human-vote leaderboard →
Agentic 3D
Scores are only comparable within a paradigm.
Rank (UB)
Generator
BT score
Votes
1
google/gemini-3.1-pro-preview (agentic)
1240.0
10
2
z-ai/glm-4.6v (agentic)
1154.4
6
3
openai/gpt-5.6-sol-pro (agentic)
1137.8
12
4
anthropic/claude-sonnet-5 (agentic)
1118.0
12
5
openai/gpt-5.6-sol (agentic)
1096.9
16
6
x-ai/grok-4.5 (agentic)
1062.8
12
7
x-ai/grok-4.20 (agentic)
996.3
12
8
qwen/qwen3.7-plus (agentic)
991.7
12
9
anthropic/claude-opus-4.8 (agentic)
984.0
10
10
minimax/minimax-m3 (agentic)
952.2
8
11
openai/gpt-5.1 (agentic)
850.4
10
12
meta-llama/llama-4-maverick (agentic)
798.5
8
13
qwen/qwen3.6-plus (agentic)
790.4
4
14
moonshotai/kimi-k2.7-code (agentic)
771.5
4
15
mistralai/mistral-medium-3-5 (agentic)
703.1
4
Image→3D reconstruction
Scores are only comparable within a paradigm.
Rank (UB)
Generator
BT score
Votes
1
Meshy 6
1223.5
12
2
Hunyuan3D v3
1186.3
16
3
Rodin/Hyper3D
1096.0
16
4
TRELLIS via Replicate
1042.7
8
5
SAM 3D
1017.4
8
6
TRELLIS via fal
1006.2
16
7
Hunyuan3D 3.1
1000.6
14
8
Hunyuan3D v2
860.8
16
9
Pixal3D
814.6
12
10
TRELLIS 2
773.7
6
LLM procedural (code-gen)
Scores are only comparable within a paradigm.
Rank (UB)
Generator
BT score
Votes
1
openai/gpt-5.6-sol
1249.8
16
2
anthropic/claude-opus-4.8
1183.6
10
3
moonshotai/kimi-k2.7-code
1165.8
14
4
z-ai/glm-5.2
1138.3
14
5
anthropic/claude-sonnet-5
1123.4
16
6
deepseek/deepseek-v3.2
1051.8
15
7
google/gemini-3.1-pro-preview
1042.1
14
8
z-ai/glm-4.6v
1026.6
14
9
deepseek/deepseek-v4-pro
1011.8
13
10
x-ai/grok-4.5
980.6
16
11
qwen/qwen3.7-plus
979.4
16
12
x-ai/grok-4.20
930.3
16
13
mistralai/mistral-medium-3-5
922.6
15
14
minimax/minimax-m3
907.7
15
15
meta-llama/llama-4-maverick
905.2
13
16
openai/gpt-5.1
863.9
14
17
qwen/qwen3.6-plus
829.4
13
Text→3D (native)
Scores are only comparable within a paradigm.
Rank (UB)
Generator
BT score
Votes
1
Tripo P1 text
1075.3
12
2
Rodin text via Replicate
1056.0
16
3
Rodin text via fal
1042.8
12
4
Hunyuan3D v3 text
1012.8
16
5
Tripo H3.1 text
984.0
12
6
Hunyuan3D 3.1 text
842.2
8
BT = Bradley–Terry score over VLM-judge ballots (multi-view condition), fitted
within a single method — scores from different methods come from disconnected
match pools and aren't comparable. Votes = judge ballots, not human votes.