P3D-Bench
Models generate parametric 3D objects from text, images, or assembly instructions. The evaluator checks whether outputs run and preserve geometry, topology, and part structure.
- Made by
- Yikang Yang and colleagues
- Released
- June 2026
- Kind
- Image generation and editing
- Level
- University
- Scored
- Checked automatically
- Answers
- Code
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Single turn
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- GPT-6 Astra #4 on the capability index
- Last model added
- September 2026
- Scores refreshed
- November 20th, 2026
GPT-6 Astra leads with 69.4 among models released by September 2026
Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. The board reports no overall score, so the headline is the plain mean of its three task scores. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
The newest model with a score here came out in September 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- Claude Opus 5.5
- and 14 more
- 1 GPT-6 Astra 69.4
- 2
Gemini 3.8 Flash 63.0
- 3 Claude Opus 5 62.6
- 4 Kimi K3 61.2
- 5 Grok 4.6 60.8
- 6
Qwen 3.8 Max 60.8
- 7 GLM-5.3-Flash 58.3
- 8
DeepSeek V4.1 Flash 58.2
- 9
Doubao Seed 2.1 Pro 49.5
Source: Official P3D-Bench leaderboard
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-6 Astra | 69.4 | Sep 2026 | Official leaderboard | |
| 2 | | 63.0 | Sep 2026 | Official leaderboard | |
| 3 | Claude Opus 5 | 62.6 | Jul 2026 | Official leaderboard | |
| 4 | Kimi K3 | 61.2 | Jul 2026 | Official leaderboard | |
| 5 | Grok 4.6 | 60.8 | Aug 2026 | Official leaderboard | |
| 6 | | 60.8 | Aug 2026 | Official leaderboard | |
| 7 | GLM-5.3-Flash | 58.3 | Aug 2026 | Official leaderboard | |
| 8 | | 58.2 | Sep 2026 | Official leaderboard | |
| 9 | | 49.5 | Jun 2026 | Official leaderboard |
Rankings by part
We read 52 parts from the benchmark’s own leaderboard in 4 groups: task, assembly-3d, text-to-3d and image-to-3d. Each ranks models on one part of the test. GPT-6 Astra leads 27 of them, and 6 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Assembly-3D
9 models we track have a score here.
- 1 GPT-6 Astra 63.5
- 2 Claude Opus 5 ↑1 54.2
- 3
Gemini 3.8 Flash ↓1 53.2
- 4 Kimi K3 50.8
- 5 Grok 4.6 50.7
- 6
Qwen 3.8 Max 49.7
- 7 GLM-5.3-Flash 48.8
- 8
DeepSeek V4.1 Flash 47.5
- 9
Doubao Seed 2.1 Pro 29.7
Text-to-3D
9 models we track have a score here.
- 1 GPT-6 Astra 84.7
- 2
Qwen 3.8 Max ↑4 82.4
- 3
Gemini 3.8 Flash ↓1 82.1
- 4 Kimi K3 81.7
- 5 Grok 4.6 81.4
- 6 Claude Opus 5 ↓3 81.1
- 7
DeepSeek V4.1 Flash ↑1 80.1
- 8
Doubao Seed 2.1 Pro ↑1 79.2
- 9 GLM-5.3-Flash ↓2 76.3
Image-to-3D
9 models we track have a score here.
- 1 GPT-6 Astra 59.9
- 2
Gemini 3.8 Flash 53.7
- 3 Claude Opus 5 52.6
- 4 Kimi K3 51.1
- 5 Grok 4.6 50.3
- 6
Qwen 3.8 Max 50.2
- 7 GLM-5.3-Flash 49.8
- 8
DeepSeek V4.1 Flash 47.0
- 9
Doubao Seed 2.1 Pro 39.5
CadQuery Geo
9 models we track have a score here.
- 1 GPT-6 Astra 68.9
- 2 Claude Opus 5 ↑1 57.1
- 3
Gemini 3.8 Flash ↓1 55.2
- 4 Grok 4.6 ↑1 53.4
- 5 GLM-5.3-Flash ↑2 52.1
- 6 Kimi K3 ↓2 51.9
- 7
Qwen 3.8 Max ↓1 49.9
- 8
DeepSeek V4.1 Flash 49.9
- 9
Doubao Seed 2.1 Pro 19.3
CadQuery Judge
9 models we track have a score here.
- 1 GPT-6 Astra 59.1
- 2 Claude Opus 5 ↑1 52.6
- 3
Gemini 3.8 Flash ↓1 52.5
- 4 Kimi K3 47.6
- 5 Grok 4.6 47.5
- 6 GLM-5.3-Flash ↑1 46.5
- 7
Qwen 3.8 Max ↓1 45.4
- 8
DeepSeek V4.1 Flash 45.4
- 9
Doubao Seed 2.1 Pro 16.4
CadQuery Part
9 models we track have a score here.
- 1 GPT-6 Astra 64.0
- 2 Claude Opus 5 ↑1 56.3
- 3 Kimi K3 ↑1 51.4
- 4
Gemini 3.8 Flash ↓2 50.3
- 5 Grok 4.6 47.3
- 6
Qwen 3.8 Max 47.2
- 7
DeepSeek V4.1 Flash ↑1 46.5
- 8 GLM-5.3-Flash ↓1 45.4
- 9
Doubao Seed 2.1 Pro 17.3
CadQuery Topo
9 models we track have a score here.
- 1 Claude Opus 5 ↑2 97.1
- 2 Grok 4.6 ↑3 95.1
- 3
Gemini 3.8 Flash ↓1 94.8
- 4 GLM-5.3-Flash ↑3 92.3
- 5
Qwen 3.8 Max ↑1 91.8
- 6 GPT-6 Astra ↓5 91.6
- 7
DeepSeek V4.1 Flash ↑1 91.3
- 8 Kimi K3 ↓4 89.8
- 9
Doubao Seed 2.1 Pro 39.3
CadQuery Valid (%)
9 models we track have a score here.
- 1 Claude Opus 5 ↑2 100%
- 2 Grok 4.6 ↑3 100%
- 3 GPT-6 Astra ↓2 98.9%
- 4 Kimi K3 97.9%
- 5
Gemini 3.8 Flash ↓3 97.9%
- 6
Qwen 3.8 Max 97.8%
- 7 GLM-5.3-Flash 97.7%
- 8
DeepSeek V4.1 Flash 96.8%
- 9
Doubao Seed 2.1 Pro 41%
OpenSCAD Geo
9 models we track have a score here.
- 1 GPT-6 Astra 67.5
- 2
Gemini 3.8 Flash 56.9
- 3 Grok 4.6 ↑2 56.6
- 4 Claude Opus 5 ↓1 55.5
- 5
Qwen 3.8 Max ↑1 54.7
- 6 Kimi K3 ↓2 53.8
- 7 GLM-5.3-Flash 53.0
- 8
DeepSeek V4.1 Flash 51.3
- 9
Doubao Seed 2.1 Pro 46.3
OpenSCAD Judge
9 models we track have a score here.
- 1 GPT-6 Astra 60.6
- 2
Gemini 3.8 Flash 53.4
- 3 Grok 4.6 ↑2 51.6
- 4 Claude Opus 5 ↓1 51.5
- 5 GLM-5.3-Flash ↑2 49.6
- 6
Qwen 3.8 Max 49.2
- 7 Kimi K3 ↓3 48.9
- 8
DeepSeek V4.1 Flash 47.3
- 9
Doubao Seed 2.1 Pro 37.9
OpenSCAD Part
9 models we track have a score here.
- 1 GPT-6 Astra 60.6
- 2 Claude Opus 5 ↑1 52.2
- 3
Qwen 3.8 Max ↑3 52.1
- 4 Kimi K3 51.0
- 5
Gemini 3.8 Flash ↓3 50.6
- 6 Grok 4.6 ↓1 47.7
- 7 GLM-5.3-Flash 46.4
- 8
DeepSeek V4.1 Flash 44.6
- 9
Doubao Seed 2.1 Pro 41.0
OpenSCAD Topo
9 models we track have a score here.
- 1 Kimi K3 ↑3 100.0
- 2 Claude Opus 5 ↑1 100.0
- 3
Qwen 3.8 Max ↑3 100.0
- 4 Grok 4.6 ↑1 100.0
- 5 GLM-5.3-Flash ↑2 100.0
- 6
Gemini 3.8 Flash ↓4 100.0
- 7 GPT-6 Astra ↓6 100.0
- 8
DeepSeek V4.1 Flash 99.0
- 9
Doubao Seed 2.1 Pro 98.0
OpenSCAD Valid (%)
9 models we track have a score here.
- 1 Kimi K3 ↑3 100%
- 2 Claude Opus 5 ↑1 100%
- 3
Qwen 3.8 Max ↑3 100%
- 4 Grok 4.6 ↑1 100%
- 5 GLM-5.3-Flash ↑2 100%
- 6
Gemini 3.8 Flash ↓4 100%
- 7 GPT-6 Astra ↓6 100%
- 8
DeepSeek V4.1 Flash 99%
- 9
Doubao Seed 2.1 Pro 98%
Average Geo
9 models we track have a score here.
- 1 GPT-6 Astra 68.2
- 2 Claude Opus 5 ↑1 56.3
- 3
Gemini 3.8 Flash ↓1 56.1
- 4 Grok 4.6 ↑1 55.0
- 5 Kimi K3 ↓1 52.8
- 6 GLM-5.3-Flash ↑1 52.6
- 7
Qwen 3.8 Max ↓1 52.3
- 8
DeepSeek V4.1 Flash 50.6
- 9
Doubao Seed 2.1 Pro 32.8
Average Judge
9 models we track have a score here.
- 1 GPT-6 Astra 59.9
- 2
Gemini 3.8 Flash 52.9
- 3 Claude Opus 5 52.0
- 4 Grok 4.6 ↑1 49.6
- 5 Kimi K3 ↓1 48.3
- 6 GLM-5.3-Flash ↑1 48.1
- 7
Qwen 3.8 Max ↓1 47.3
- 8
DeepSeek V4.1 Flash 46.3
- 9
Doubao Seed 2.1 Pro 27.1
Average Part
9 models we track have a score here.
- 1 GPT-6 Astra 62.3
- 2 Claude Opus 5 ↑1 54.3
- 3 Kimi K3 ↑1 51.2
- 4
Gemini 3.8 Flash ↓2 50.5
- 5
Qwen 3.8 Max ↑1 49.6
- 6 Grok 4.6 ↓1 47.5
- 7 GLM-5.3-Flash 45.9
- 8
DeepSeek V4.1 Flash 45.6
- 9
Doubao Seed 2.1 Pro 29.2
Average Topo
9 models we track have a score here.
- 1 Claude Opus 5 ↑2 98.5
- 2 Grok 4.6 ↑3 97.6
- 3
Gemini 3.8 Flash ↓1 97.4
- 4 GLM-5.3-Flash ↑3 96.2
- 5
Qwen 3.8 Max ↑1 95.9
- 6 GPT-6 Astra ↓5 95.8
- 7
DeepSeek V4.1 Flash ↑1 95.1
- 8 Kimi K3 ↓4 94.9
- 9
Doubao Seed 2.1 Pro 68.6
Average Valid (%)
9 models we track have a score here.
- 1 Claude Opus 5 ↑2 100%
- 2 Grok 4.6 ↑3 100%
- 3 GPT-6 Astra ↓2 99.4%
- 4 Kimi K3 99%
- 5
Gemini 3.8 Flash ↓3 99%
- 6
Qwen 3.8 Max 98.9%
- 7 GLM-5.3-Flash 98.9%
- 8
DeepSeek V4.1 Flash 97.9%
- 9
Doubao Seed 2.1 Pro 69.5%
Descriptive JSON Judge
9 models we track have a score here.
- 1 GPT-6 Astra 92.7
- 2
Qwen 3.8 Max ↑4 86.7
- 3
Gemini 3.8 Flash ↓1 85.6
- 4 Grok 4.6 ↑1 84.3
- 5
DeepSeek V4.1 Flash ↑3 84.1
- 6 Kimi K3 ↓2 83.1
- 7 GLM-5.3-Flash 83.1
- 8 Claude Opus 5 ↓5 82.9
- 9
Doubao Seed 2.1 Pro 82.6
Descriptive JSON Valid (%)
9 models we track have a score here.
- 1 Claude Opus 5 ↑2 100%
- 2
Qwen 3.8 Max ↑4 100%
- 3 Grok 4.6 ↑2 100%
- 4
Gemini 3.8 Flash ↓2 100%
- 5 GPT-6 Astra ↓4 100%
- 6
DeepSeek V4.1 Flash ↑2 100%
- 7
Doubao Seed 2.1 Pro ↑2 99%
- 8 GLM-5.3-Flash ↓1 98%
- 9 Kimi K3 ↓5 97%
Descriptive OpenSCAD Judge
9 models we track have a score here.
- 1 GPT-6 Astra 92.6
- 2 Kimi K3 ↑2 90.8
- 3
Qwen 3.8 Max ↑3 90.5
- 4
DeepSeek V4.1 Flash ↑4 88.4
- 5
Gemini 3.8 Flash ↓3 87.6
- 6 Grok 4.6 ↓1 87.5
- 7 Claude Opus 5 ↓4 86.3
- 8
Doubao Seed 2.1 Pro ↑1 85.9
- 9 GLM-5.3-Flash ↓2 79.1
Descriptive OpenSCAD Valid (%)
9 models we track have a score here.
- 1
Doubao Seed 2.1 Pro ↑8 100%
- 2 Kimi K3 ↑2 100%
- 3 Claude Opus 5 100%
- 4
Qwen 3.8 Max ↑2 100%
- 5 Grok 4.6 100%
- 6
Gemini 3.8 Flash ↓4 100%
- 7 GPT-6 Astra ↓6 100%
- 8
DeepSeek V4.1 Flash 100%
- 9 GLM-5.3-Flash ↓2 97%
Descriptive Average Judge
9 models we track have a score here.
- 1 GPT-6 Astra 92.6
- 2
Qwen 3.8 Max ↑4 88.6
- 3 Kimi K3 ↑1 86.9
- 4
Gemini 3.8 Flash ↓2 86.6
- 5
DeepSeek V4.1 Flash ↑3 86.2
- 6 Grok 4.6 ↓1 85.9
- 7 Claude Opus 5 ↓4 84.6
- 8
Doubao Seed 2.1 Pro ↑1 84.2
- 9 GLM-5.3-Flash ↓2 81.1
Descriptive Average Valid (%)
9 models we track have a score here.
- 1 Claude Opus 5 ↑2 100%
- 2
Qwen 3.8 Max ↑4 100%
- 3 Grok 4.6 ↑2 100%
- 4
Gemini 3.8 Flash ↓2 100%
- 5 GPT-6 Astra ↓4 100%
- 6
DeepSeek V4.1 Flash ↑2 100%
- 7
Doubao Seed 2.1 Pro ↑2 99.5%
- 8 Kimi K3 ↓4 98.5%
- 9 GLM-5.3-Flash ↓2 97.5%
Parametric JSON Geo
9 models we track have a score here.
- 1 Grok 4.6 ↑4 68.1
- 2
Gemini 3.8 Flash 67.8
- 3
Qwen 3.8 Max ↑3 67.0
- 4 GPT-6 Astra ↓3 66.5
- 5 Claude Opus 5 ↓2 65.9
- 6 Kimi K3 ↓2 65.6
- 7
DeepSeek V4.1 Flash ↑1 64.9
- 8
Doubao Seed 2.1 Pro ↑1 64.0
- 9 GLM-5.3-Flash ↓2 61.2
Parametric JSON Judge
9 models we track have a score here.
- 1 GPT-6 Astra 91.1
- 2
Qwen 3.8 Max ↑4 89.6
- 3 Claude Opus 5 88.6
- 4 Kimi K3 88.5
- 5
Gemini 3.8 Flash ↓3 88.1
- 6 Grok 4.6 ↓1 87.8
- 7
DeepSeek V4.1 Flash ↑1 86.4
- 8
Doubao Seed 2.1 Pro ↑1 86.3
- 9 GLM-5.3-Flash ↓2 85.8
Parametric JSON Topo
9 models we track have a score here.
- 1 Claude Opus 5 ↑2 99.0
- 2 Grok 4.6 ↑3 99.0
- 3 GPT-6 Astra ↓2 98.7
- 4
Gemini 3.8 Flash ↓2 98.6
- 5
Qwen 3.8 Max ↑1 97.9
- 6
Doubao Seed 2.1 Pro ↑3 97.6
- 7 Kimi K3 ↓3 97.6
- 8
DeepSeek V4.1 Flash 97.6
- 9 GLM-5.3-Flash ↓2 96.3
Parametric JSON Valid (%)
9 models we track have a score here.
- 1 Claude Opus 5 ↑2 100%
- 2
Qwen 3.8 Max ↑4 100%
- 3 Grok 4.6 ↑2 100%
- 4
Gemini 3.8 Flash ↓2 100%
- 5 GPT-6 Astra ↓4 100%
- 6
Doubao Seed 2.1 Pro ↑3 99%
- 7 Kimi K3 ↓3 99%
- 8
DeepSeek V4.1 Flash 99%
- 9 GLM-5.3-Flash ↓2 98%
Parametric OpenSCAD Geo
9 models we track have a score here.
- 1 GPT-6 Astra 71.0
- 2
Gemini 3.8 Flash 70.5
- 3 Claude Opus 5 69.3
- 4 Grok 4.6 ↑1 69.3
- 5
Qwen 3.8 Max ↑1 68.7
- 6 Kimi K3 ↓2 68.4
- 7
DeepSeek V4.1 Flash ↑1 66.1
- 8
Doubao Seed 2.1 Pro ↑1 65.3
- 9 GLM-5.3-Flash ↓2 60.3
Parametric OpenSCAD Judge
9 models we track have a score here.
- 1 GPT-6 Astra 94.4
- 2 Kimi K3 ↑2 94.0
- 3 Claude Opus 5 93.8
- 4
Gemini 3.8 Flash ↓2 93.0
- 5
Qwen 3.8 Max ↑1 91.8
- 6 Grok 4.6 ↓1 91.4
- 7
Doubao Seed 2.1 Pro ↑2 91.0
- 8
DeepSeek V4.1 Flash 90.8
- 9 GLM-5.3-Flash ↓2 88.7
Parametric OpenSCAD Topo
9 models we track have a score here.
- 1
Doubao Seed 2.1 Pro ↑8 100.0
- 2 Kimi K3 ↑2 100.0
- 3 Claude Opus 5 100.0
- 4
Qwen 3.8 Max ↑2 100.0
- 5 Grok 4.6 100.0
- 6
Gemini 3.8 Flash ↓4 100.0
- 7 GPT-6 Astra ↓6 100.0
- 8 GLM-5.3-Flash ↓1 99.0
- 9
DeepSeek V4.1 Flash ↓1 99.0
Parametric OpenSCAD Valid (%)
9 models we track have a score here.
- 1
Doubao Seed 2.1 Pro ↑8 100%
- 2 Kimi K3 ↑2 100%
- 3 Claude Opus 5 100%
- 4
Qwen 3.8 Max ↑2 100%
- 5 Grok 4.6 100%
- 6
Gemini 3.8 Flash ↓4 100%
- 7 GPT-6 Astra ↓6 100%
- 8 GLM-5.3-Flash ↓1 99%
- 9
DeepSeek V4.1 Flash ↓1 99%
Parametric Average Geo
9 models we track have a score here.
- 1
Gemini 3.8 Flash ↑1 69.2
- 2 GPT-6 Astra ↓1 68.8
- 3 Grok 4.6 ↑2 68.7
- 4
Qwen 3.8 Max ↑2 67.9
- 5 Claude Opus 5 ↓2 67.6
- 6 Kimi K3 ↓2 67.0
- 7
DeepSeek V4.1 Flash ↑1 65.5
- 8
Doubao Seed 2.1 Pro ↑1 64.6
- 9 GLM-5.3-Flash ↓2 60.8
Parametric Average Judge
9 models we track have a score here.
- 1 GPT-6 Astra 92.8
- 2 Kimi K3 ↑2 91.3
- 3 Claude Opus 5 91.2
- 4
Qwen 3.8 Max ↑2 90.7
- 5
Gemini 3.8 Flash ↓3 90.5
- 6 Grok 4.6 ↓1 89.6
- 7
Doubao Seed 2.1 Pro ↑2 88.7
- 8
DeepSeek V4.1 Flash 88.6
- 9 GLM-5.3-Flash ↓2 87.2
Parametric Average Topo
9 models we track have a score here.
- 1 Claude Opus 5 ↑2 99.5
- 2 Grok 4.6 ↑3 99.5
- 3
Gemini 3.8 Flash ↓1 99.3
- 4 GPT-6 Astra ↓3 99.3
- 5
Qwen 3.8 Max ↑1 99.0
- 6
Doubao Seed 2.1 Pro ↑3 98.8
- 7 Kimi K3 ↓3 98.8
- 8
DeepSeek V4.1 Flash 98.3
- 9 GLM-5.3-Flash ↓2 97.6
Parametric Average Valid (%)
9 models we track have a score here.
- 1 Claude Opus 5 ↑2 100%
- 2
Qwen 3.8 Max ↑4 100%
- 3 Grok 4.6 ↑2 100%
- 4
Gemini 3.8 Flash ↓2 100%
- 5 GPT-6 Astra ↓4 100%
- 6
Doubao Seed 2.1 Pro ↑3 99.5%
- 7 Kimi K3 ↓3 99.5%
- 8
DeepSeek V4.1 Flash 99%
- 9 GLM-5.3-Flash ↓2 98.5%
CadQuery Geo
9 models we track have a score here.
- 1 GPT-6 Astra 58.1
- 2
Gemini 3.8 Flash 52.8
- 3 Claude Opus 5 50.7
- 4 Kimi K3 50.0
- 5 GLM-5.3-Flash ↑2 49.6
- 6
Qwen 3.8 Max 49.5
- 7 Grok 4.6 ↓2 46.8
- 8
DeepSeek V4.1 Flash 42.5
- 9
Doubao Seed 2.1 Pro 24.6
CadQuery Judge
9 models we track have a score here.
- 1 GPT-6 Astra 59.9
- 2
Gemini 3.8 Flash 53.7
- 3 Claude Opus 5 51.3
- 4 Kimi K3 50.7
- 5
Qwen 3.8 Max ↑1 49.7
- 6 GLM-5.3-Flash ↑1 48.6
- 7 Grok 4.6 ↓2 47.5
- 8
DeepSeek V4.1 Flash 40.9
- 9
Doubao Seed 2.1 Pro 24.4
CadQuery Topo
9 models we track have a score here.
- 1 Kimi K3 ↑3 98.1
- 2 GLM-5.3-Flash ↑5 97.6
- 3
Qwen 3.8 Max ↑3 96.5
- 4
Gemini 3.8 Flash ↓2 96.1
- 5 Claude Opus 5 ↓2 96.0
- 6 Grok 4.6 ↓1 94.5
- 7 GPT-6 Astra ↓6 92.9
- 8
DeepSeek V4.1 Flash 89.1
- 9
Doubao Seed 2.1 Pro 52.0
CadQuery Valid (%)
9 models we track have a score here.
- 1 Kimi K3 ↑3 100%
- 2 Claude Opus 5 ↑1 100%
- 3 GLM-5.3-Flash ↑4 100%
- 4
Qwen 3.8 Max ↑2 99%
- 5 GPT-6 Astra ↓4 98.8%
- 6
Gemini 3.8 Flash ↓4 98%
- 7 Grok 4.6 ↓2 97%
- 8
DeepSeek V4.1 Flash 91%
- 9
Doubao Seed 2.1 Pro 55%
OpenSCAD Geo
9 models we track have a score here.
- 1 GPT-6 Astra 57.0
- 2
Gemini 3.8 Flash 52.6
- 3 Claude Opus 5 51.5
- 4 Kimi K3 50.4
- 5 Grok 4.6 50.2
- 6
Qwen 3.8 Max 49.3
- 7
DeepSeek V4.1 Flash ↑1 48.8
- 8 GLM-5.3-Flash ↓1 47.8
- 9
Doubao Seed 2.1 Pro 46.2
OpenSCAD Judge
9 models we track have a score here.
- 1 GPT-6 Astra 60.7
- 2
Gemini 3.8 Flash 55.2
- 3 Claude Opus 5 53.4
- 4 Kimi K3 52.5
- 5 Grok 4.6 52.1
- 6
DeepSeek V4.1 Flash ↑2 50.2
- 7
Qwen 3.8 Max ↓1 50.1
- 8 GLM-5.3-Flash ↓1 48.8
- 9
Doubao Seed 2.1 Pro 43.9
OpenSCAD Topo
9 models we track have a score here.
- 1 Kimi K3 ↑3 100.0
- 2 Claude Opus 5 ↑1 100.0
- 3
Qwen 3.8 Max ↑3 100.0
- 4 Grok 4.6 ↑1 100.0
- 5
Gemini 3.8 Flash ↓3 100.0
- 6 GPT-6 Astra ↓5 100.0
- 7
DeepSeek V4.1 Flash ↑1 100.0
- 8
Doubao Seed 2.1 Pro ↑1 99.0
- 9 GLM-5.3-Flash ↓2 97.9
OpenSCAD Valid (%)
9 models we track have a score here.
- 1 Kimi K3 ↑3 100%
- 2 Claude Opus 5 ↑1 100%
- 3
Qwen 3.8 Max ↑3 100%
- 4 Grok 4.6 ↑1 100%
- 5
Gemini 3.8 Flash ↓3 100%
- 6 GPT-6 Astra ↓5 100%
- 7
DeepSeek V4.1 Flash ↑1 100%
- 8
Doubao Seed 2.1 Pro ↑1 99%
- 9 GLM-5.3-Flash ↓2 98%
Three.js Geo
9 models we track have a score here.
- 1 GPT-6 Astra 61.5
- 2 Claude Opus 5 ↑1 54.2
- 3
Gemini 3.8 Flash ↓1 53.8
- 4 Grok 4.6 ↑1 53.1
- 5
Qwen 3.8 Max ↑1 51.9
- 6 Kimi K3 ↓2 51.8
- 7 GLM-5.3-Flash 51.6
- 8
Doubao Seed 2.1 Pro ↑1 51.1
- 9
DeepSeek V4.1 Flash ↓1 50.9
Three.js Judge
9 models we track have a score here.
- 1 GPT-6 Astra 62.2
- 2 Claude Opus 5 ↑1 54.6
- 3
Gemini 3.8 Flash ↓1 54.1
- 4 GLM-5.3-Flash ↑3 52.2
- 5 Grok 4.6 52.1
- 6 Kimi K3 ↓2 51.3
- 7
Qwen 3.8 Max ↓1 50.9
- 8
DeepSeek V4.1 Flash 48.7
- 9
Doubao Seed 2.1 Pro 47.0
Three.js Topo
9 models we track have a score here.
- 1
Qwen 3.8 Max ↑5 86.6
- 2
DeepSeek V4.1 Flash ↑6 84.4
- 3 GPT-6 Astra ↓2 83.8
- 4 Claude Opus 5 ↓1 81.2
- 5 GLM-5.3-Flash ↑2 80.7
- 6 Kimi K3 ↓2 79.4
- 7
Doubao Seed 2.1 Pro ↑2 78.5
- 8 Grok 4.6 ↓3 77.0
- 9
Gemini 3.8 Flash ↓7 71.0
Three.js Valid (%)
9 models we track have a score here.
- 1
Doubao Seed 2.1 Pro ↑8 100%
- 2 Kimi K3 ↑2 100%
- 3 Claude Opus 5 100%
- 4
Qwen 3.8 Max ↑2 100%
- 5 Grok 4.6 100%
- 6 GLM-5.3-Flash ↑1 100%
- 7 GPT-6 Astra ↓6 100%
- 8
DeepSeek V4.1 Flash 100%
- 9
Gemini 3.8 Flash ↓7 99%
Average Geo
9 models we track have a score here.
- 1 GPT-6 Astra 58.9
- 2
Gemini 3.8 Flash 53.1
- 3 Claude Opus 5 52.1
- 4 Kimi K3 50.7
- 5
Qwen 3.8 Max ↑1 50.3
- 6 Grok 4.6 ↓1 50.0
- 7 GLM-5.3-Flash 49.7
- 8
DeepSeek V4.1 Flash 47.4
- 9
Doubao Seed 2.1 Pro 40.6
Average Judge
9 models we track have a score here.
- 1 GPT-6 Astra 60.9
- 2
Gemini 3.8 Flash 54.3
- 3 Claude Opus 5 53.1
- 4 Kimi K3 51.5
- 5 Grok 4.6 50.6
- 6
Qwen 3.8 Max 50.2
- 7 GLM-5.3-Flash 49.9
- 8
DeepSeek V4.1 Flash 46.6
- 9
Doubao Seed 2.1 Pro 38.4
Average Topo
9 models we track have a score here.
- 1
Qwen 3.8 Max ↑5 94.4
- 2 Kimi K3 ↑2 92.5
- 3 Claude Opus 5 92.4
- 4 GPT-6 Astra ↓3 92.2
- 5 GLM-5.3-Flash ↑2 92.1
- 6
DeepSeek V4.1 Flash ↑2 91.2
- 7 Grok 4.6 ↓2 90.5
- 8
Gemini 3.8 Flash ↓6 89.0
- 9
Doubao Seed 2.1 Pro 76.5
Average Valid (%)
9 models we track have a score here.
- 1 Kimi K3 ↑3 100%
- 2 Claude Opus 5 ↑1 100%
- 3
Qwen 3.8 Max ↑3 99.7%
- 4 GPT-6 Astra ↓3 99.6%
- 5 GLM-5.3-Flash ↑2 99.3%
- 6 Grok 4.6 ↓1 99%
- 7
Gemini 3.8 Flash ↓5 99%
- 8
DeepSeek V4.1 Flash 97%
- 9
Doubao Seed 2.1 Pro 84.7%
The best score, 69.4, costs $0.89 a task
Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: Official P3D-Bench leaderboard
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| GPT-6 Astra | 69.4 | $0.89 | |||
| 63.0 | $0.12 | ||||
| Claude Opus 5 | 62.6 | $0.59 | |||
| Kimi K3 | 61.2 | $0.38 | |||
| Grok 4.6 | 60.8 | $0.19 | |||
| 60.8 | $0.10 | ||||
| GLM-5.3-Flash | 58.3 | $0.02 | |||
| 58.2 | $0.04 | ||||
| 49.5 | $0.22 |
5 models have held number one on the official leaderboard since its release in June 2026
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official P3D-Bench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official P3D-Bench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Vision
- MMMU In use 11.5K college-level multimodal questions across 30 subjects. Multiple-choice and open accuracy.
Qwen 3.6 Plus 86%
- MMMU-Pro In use Harder MMMU: college-level multimodal questions with more options and vision-only inputs to remove text shortcuts. GPT-5.6 Sol 84.6%
- CharXiv In use Chart understanding on real arXiv figures. Reasoning questions require synthesizing information from complex charts. Claude Fable 5 93.5%
- MathVista In use Math reasoning in visual contexts (charts, geometry, puzzles) from 31 datasets. Answer accuracy.
Seed2.1 Pro 90.7%
- MATH-Vision (MathVision) In use 3,040 competition math problems with visual context. Answer accuracy.
Qwen 3.8 Max 97.7%
- MMStar In use Models answer 1,500 human-curated image questions that cannot be solved without the image, across 6 core capabilities. Scored by accuracy.
Qwen 3.8 Max 85.9%
Data audit
All 9 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 3 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (9 entries).
- Pass Models scored 9 models we track.
- Pass Newest score The newest is from September 2026, for DeepSeek V4.1 Flash.
- Note Frontier models Only 1 of the top 10 has a score. The strongest tested is GPT-6 Astra (#4).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on spatiaos.github.io: 9 rows for models we track at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.