Rosa in 3D
How well do generative 3D models rebuild Rosa? Every output below was produced by a model from the same brief, then compared blind against reference photographs of the real organism — so the question is biological fidelity, not which mesh looks prettiest.
31
3D outputs
31
Models attempted
93
Human votes
🌿
Plants
The real organism
CC-licensed reference photographs
Models that attempted it
grouped by generation method| Model | Method | Outputs | Votes |
|---|---|---|---|
| SAM 3D | Image→3D reconstruction | 1 | 21 |
| Meshy 6 | Image→3D reconstruction | 1 | 20 |
| google/gemini-3.1-pro-preview | LLM procedural (code-gen) | 1 | 2 |
| anthropic/claude-opus-4.8 | LLM procedural (code-gen) | 1 | 1 |
| anthropic/claude-sonnet-5 | LLM procedural (code-gen) | 1 | 1 |
| deepseek/deepseek-v3.2 | LLM procedural (code-gen) | 1 | 1 |
| deepseek/deepseek-v4-pro | LLM procedural (code-gen) | 1 | 1 |
| mistralai/mistral-medium-3-5 | LLM procedural (code-gen) | 1 | 1 |
| openai/gpt-5.1 | LLM procedural (code-gen) | 1 | 1 |
| openai/gpt-5.6-sol | LLM procedural (code-gen) | 1 | 1 |
| qwen/qwen3.6-plus | LLM procedural (code-gen) | 1 | 1 |
| x-ai/grok-4.20 | LLM procedural (code-gen) | 1 | 1 |
| x-ai/grok-4.3 | LLM procedural (code-gen) | 1 | 1 |
| z-ai/glm-4.6v | LLM procedural (code-gen) | 1 | 1 |
| z-ai/glm-5.2 | LLM procedural (code-gen) | 1 | 1 |
| minimax/minimax-m3 | LLM procedural (code-gen) | 1 | 0 |
| qwen/qwen3.7-plus | LLM procedural (code-gen) | 1 | 0 |
| x-ai/grok-4.5 | LLM procedural (code-gen) | 1 | 0 |
| Hunyuan3D 3.1 text | Text→3D (native) | 1 | 20 |
| Hunyuan3D v3 text | Text→3D (native) | 1 | 20 |
| Meshy v6 text | Text→3D (native) | 1 | 19 |
| Rodin text via fal | Text→3D (native) | 1 | 19 |
| Tripo H3.1 text | Text→3D (native) | 1 | 19 |
| Tripo P1 text | Text→3D (native) | 1 | 19 |
| anthropic/claude-opus-4.8 (agentic) | Agentic 3D | 1 | 2 |
| google/gemini-3.1-pro-preview (agentic) | Agentic 3D | 1 | 1 |
| qwen/qwen3.7-plus (agentic) | Agentic 3D | 1 | 1 |
| x-ai/grok-4.20 (agentic) | Agentic 3D | 1 | 1 |
| z-ai/glm-4.6v (agentic) | Agentic 3D | 1 | 1 |
| openai/gpt-5.6-sol (agentic) | Agentic 3D | 1 | 0 |
| x-ai/grok-4.5 (agentic) | Agentic 3D | 1 | 0 |
Models are ranked within their own method, not against each other across methods — see the leaderboards for standings.
Tasks
| Task | Tier | Outputs | Votes |
|---|---|---|---|
| Rosa — single-image → 3D reconstruction | hard | 31 | 93 |