Rapidata/text-to-3d-with-llm-code

Rapidata Text-to-3D with LLM code Benchmark

Built by Rapidata, collected with the Rapidata Python SDK.

To evaluate your own models and create a leaderboard, check out our MRI.

Live Leaderboard

Explore the full interactive leaderboard — filter by model, question and prompt, and inspect individual head-to-head matchups on Rapidata.

Open the live leaderboard

At a glance

Head-to-head comparisons (rows)	44,482
Human votes	780,424
Models compared	14
Prompts	500

The leaderboards

Leaderboard	Question shown to annotators	Prompt shown?
Creativity	Which object is more creative and interesting?	No
Coherence	Which 3D model has more errors, like floating, misplaced or broken parts?	No
Alignment	Which 3D model matches the description better?	Yes
Preference	Which 3D model looks better?	No

Examples

The most decisive matchups on each leaderboard, picked the same way as the Matchups tab on benchmark.ai: the split must hold at 1.5 standard errors or more, and the strongest come first. No prompt appears twice, and each leaderboard covers a different mix of categories and winning models.

Preference

Which 3D model looks better?

Prompt	Winner	Loser
A red and white striped circus tent with a flag on top	gpt-6-astra 5 of 5 votes	deepseek-v4-pro
A tugboat with tires hanging along its sides	gemini-3.8-flash 5 of 5 votes	deepseek-v4-pro
A refrigerator with its door open	claude-opus-5.5 12 of 12 votes	kimi-k3

Alignment

Which 3D model matches the description better?

The only question where annotators also saw the prompt.

Prompt	Winner	Loser
A butterfly with blue wings	gpt-6-astra 12 of 12 votes	grok-4.7
A flower pot with a trowel stuck in it	claude-opus-5.5 5 of 5 votes	deepseek-v4-pro
A pumpkin carriage from a fairy tale	gpt-6.1-sol 5 of 5 votes	mistral-large-4

Coherence

Which 3D model has more errors, like floating, misplaced or broken parts?

Annotators picked the object with more errors, so the winner shown is the one they did not pick.

Prompt	Winner	Loser
An octopus with eight curled tentacles	gpt-6-astra 12 of 12 votes	kimi-k3
A wheelbarrow full of bricks	gpt-6.1-sol 5 of 5 votes	kimi-k3
A doghouse with a name sign above the door	claude-opus-5.5 10 of 12 votes	gemini-3.8-flash

Creativity

Which object is more creative and interesting?

Prompt	Winner	Loser
A windmill with four sails	gpt-6.1-sol 5 of 5 votes	mistral-large-4
A loft bed with a desk underneath	muse-wonder 10 of 10 votes	qwen3.8-max-prime
A starfish with five arms	claude-opus-5.5 10 of 10 votes	deepseek-v4-pro

Dataset structure

One row per head-to-head comparison of two samples generated from the same prompt.

Column	Type	Meaning
prompt	string	The prompt the two compared samples were generated from.
video1	string	Compared video 1: public URL of the video.
video2	string	Compared video 2: public URL of the video.
model1	string	Name of the model that produced video1.
model2	string	Name of the model that produced video2.
weighted_results_video1_creativity	float32	Sum of the userScore weights of the votes for video1 on the Creativity leaderboard.
weighted_results_video2_creativity	float32	Sum of the userScore weights of the votes for video2 on the Creativity leaderboard.
detailed_results_creativity	string	Every vote on the Creativity leaderboard as a JSON array; votedFor is A for side 1 and B for side 2.
weighted_results_video1_coherence	float32	Sum of the userScore weights of the votes for video1 on the Coherence leaderboard.
weighted_results_video2_coherence	float32	Sum of the userScore weights of the votes for video2 on the Coherence leaderboard.
detailed_results_coherence	string	Every vote on the Coherence leaderboard as a JSON array; votedFor is A for side 1 and B for side 2.
weighted_results_video1_alignment	float32	Sum of the userScore weights of the votes for video1 on the Alignment leaderboard.
weighted_results_video2_alignment	float32	Sum of the userScore weights of the votes for video2 on the Alignment leaderboard.
detailed_results_alignment	string	Every vote on the Alignment leaderboard as a JSON array; votedFor is A for side 1 and B for side 2.
weighted_results_video1_preference	float32	Sum of the userScore weights of the votes for video1 on the Preference leaderboard.
weighted_results_video2_preference	float32	Sum of the userScore weights of the votes for video2 on the Preference leaderboard.
detailed_results_preference	string	Every vote on the Preference leaderboard as a JSON array; votedFor is A for side 1 and B for side 2.

The weighted_results_* pair is a userScore-weighted vote sum per side, not a probability: the two values of a row do not sum to 1. Divide by their sum for a normalised win share.

Overall ranking

#	Model	ELO
1	gemini-4-argon	1581.55
2	gpt-6-astra	1578.51
3	gpt-6.1-sol	1460.51
4	claude-opus-5.5	1316.63
5	muse-wonder	1311.88
6	gemini-3.8-flash	1124.86
7	claude-fable-5.1	1098.67
8	grok-4.7	991.13
9	kimi-k3	742.366
10	qwen3.8-max-prime	701.778
11	glm-5.3-prime	670.075
12	glm-5.3-flash	559.467
13	mistral-large-4	492.705
14	deepseek-v4-pro	369.868

Creativity ranking

#	Model	ELO
1	gemini-4-argon	1794.94
2	gpt-6-astra	1423.91
3	gemini-3.8-flash	1322.04
4	gpt-6.1-sol	1299.32
5	muse-wonder	1274.79
6	claude-opus-5.5	1163.15
7	grok-4.7	1052.69
8	claude-fable-5.1	982.764
9	glm-5.3-prime	766.745
10	kimi-k3	760.325
11	qwen3.8-max-prime	717.383
12	glm-5.3-flash	637.826
13	mistral-large-4	462.044
14	deepseek-v4-pro	342.073

Coherence ranking

#	Model	ELO
1	gpt-6-astra	1619.89
2	gpt-6.1-sol	1499.65
3	claude-opus-5.5	1388.76
4	muse-wonder	1290.42
5	gemini-4-argon	1271.85
6	claude-fable-5.1	1164.81
7	grok-4.7	967.768
8	gemini-3.8-flash	912.763
9	kimi-k3	770.67
10	qwen3.8-max-prime	748.593
11	glm-5.3-prime	643.177
12	mistral-large-4	627.273
13	glm-5.3-flash	568.274
14	deepseek-v4-pro	526.101

Alignment ranking

#	Model	ELO
1	gpt-6-astra	1601.59
2	gemini-4-argon	1576.04
3	gpt-6.1-sol	1500.92
4	claude-opus-5.5	1357.94
5	muse-wonder	1331.09
6	claude-fable-5.1	1146.26
7	gemini-3.8-flash	1052.44
8	grok-4.7	976.087
9	kimi-k3	766.661
10	qwen3.8-max-prime	704.244
11	glm-5.3-prime	661.558
12	glm-5.3-flash	492.166
13	mistral-large-4	478.849
14	deepseek-v4-pro	354.162

Preference ranking

#	Model	ELO
1	gemini-4-argon	1794.09
2	gpt-6-astra	1690.55
3	gpt-6.1-sol	1552.64
4	muse-wonder	1439.38
5	claude-opus-5.5	1357.33
6	gemini-3.8-flash	1186.73
7	claude-fable-5.1	1084.72
8	grok-4.7	993.241
9	kimi-k3	666.817
10	qwen3.8-max-prime	629.546
11	glm-5.3-prime	582.313
12	glm-5.3-flash	447.393
13	mistral-large-4	377.978
14	deepseek-v4-pro	197.273

Citation

If you use this dataset, please cite it:

@misc{giger2026textto3d,
  title     = {{text-to-3d-with-llm-code}},
  author    = {Giger, Lino},
  year      = {2026},
  publisher = {Hugging Face},
  doi       = {10.57967/hf/11174},
  url       = {https://huggingface.co/datasets/Rapidata/text-to-3d-with-llm-code}
}

Giger, L. (2026). text-to-3d-with-llm-code [Data set]. Hugging Face. https://doi.org/10.57967/hf/11174

About Rapidata

Rapidata's technology makes collecting human feedback at scale faster and more accessible than ever before. Visit rapidata.ai to learn more.