yupp-ai/yupp-svg-20251204

---
license: cc-by-4.0
---

# Yupp SVG Dataset: Exploration of the Reasoning and Coding Abilities of Frontier Models

## 1. Overview

This dataset contains organic user preferences collected from [Yupp](https://yupp.ai), where users compare side-by-side AI-generated SVG outputs.
Each entry represents a user chat with model responses and preference comparisons, providing a direct evaluation of model reasoning and coding capabilities through the lens of SVG generation.
This release represents only a small fraction of the SVG data Yupp has gathered; it is an initial public slice intended to support research.

Generation of SVG (Scalable Vector Graphics) is interesting in that it exercises a model's code generation abilities.
It subtly reveals aspects of a model's "understanding" of the world and objects within it, for example:
- Models need to have internal representations of objects that capture hierarchical and spatial relationships.
- Models need to demonstrate basic spatial reasoning abilities.
- Models need to capture simple physics as a prerequisite for animations that make sense.
- Models need to express a sense of aesthetics, which parallels style and tone in text generation.

For more details, see our [blog post](https://yupp.ai/blog/svg).

## 2. Dataset Overview

### 2.1 Basic Statistics

| Metric | Count |
|--------|------:|
| Total Chats | 2,816 |
| Total Turns | 3,527 |
| Single-turn Chats | 2,227 |
| Multi-turn Chats | 589 |
| Total Preferences | 3,750 |
| Unique Models | 22 |
| Date Range | Oct 12, 2025 – Dec 2, 2025 |

### 2.2 Language Distribution

| Language | Count |
|----------|------:|
| English | 3,024 |
| Spanish | 237 |
| Portuguese | 146 |
| French | 39 |
| Greek | 38 |
| Chinese | 14 |
| Russian | 9 |
| Indonesian | 7 |
| Arabic | 6 |
| Others | 7 |

### 2.3 Models Included

The dataset features responses from 22 frontier AI models:

- **Alibaba**: Qwen3 Max Instruct Preview, Qwen3 Max Thinking Preview
- **Anthropic**: Claude Opus 4.5 (Thinking), Claude Sonnet 4.5, Claude Sonnet 4.5 (Thinking), Claude Haiku 4.5
- **DeepSeek**: DeepSeek V3.2 Exp Thinking
- **Google**: Gemini 2.5 Pro High, Gemini 3 Pro
- **MiniMax**: MiniMax M2
- **Mistral**: Mistral Medium 3.1
- **Moonshot**: Kimi K2 0905, Kimi K2 Thinking Turbo
- **OpenAI**: GPT-5 (High), GPT-5 Chat, GPT-5 Codex (High), GPT-5.1 (High), GPT-5.1 Codex (High), gpt-oss-120b
- **xAI**: Grok 4 Fast Reasoning, Grok 4.1 Fast Reasoning
- **Zhipu AI**: GLM 4.6

## 3. Data Schema and Key Fields

The dataset is structured as a JSON file with the following schema:

```json
{
"chats": [
{
"chat_id": "string (anonymized unique identifier)",
"turns": [
{
"type": "text",
"turn_sequence_id": 0,
"turn_id": "string (anonymized unique identifier)",
"prompt": "string (user prompt)",
"created_at": "ISO 8601 timestamp",
"language_code": "string (e.g., 'en', 'es', 'pt')",
"responses": [
{
"name": "string (model name)",
"content": "string (model response with SVG code)"
}
],
"preferences": [
{
"type": "comparison",
"liked": {
"name": "string (winning model name)",
"notes": ["TRAIT_1", "TRAIT_2"]
},
"disliked": [
{
"name": "string (losing model name)",
"notes": ["TRAIT_1", "TRAIT_2"]
}
],
"user_comment": "string (optional free-text feedback)"
}
]
}
]
}
]
}
```

### 3.1 Key Fields

| Field | Description |
|-------|-------------|
| `chat_id` | Anonymized unique identifier for each chat |
| `turns` | Array of turns (for multi-turn chats) |
| `turn_sequence_id` | Sequential index of the turn within the chat |
| `prompt` | The user's prompt for SVG generation |
| `language_code` | ISO language code of the prompt |
| `responses` | Array of model responses (typically two) |
| `preferences` | User preference judgments with traits and optional comments |

### 3.2 Preference Types

- **`comparison`**: One model is preferred over another model (or more than one model). Contains both `liked` and `disliked` fields.
- **`downvote`**: The model response is unsatisfactory. Only contains `disliked` field.

## 4. Citation and Contact

If you use this dataset in your research, please cite:

```bibtex
@dataset{yupp_svg_2025,
title={Yupp SVG Dataset: Exploration of the Reasoning and Coding Abilities of Frontier Models},
author={Yupp AI},
year={2025},
url={https://huggingface.co/datasets/yupp/yupp-svg-20251204}
}
```

For questions, collaborations, or feedback, reach out to us at [[e-mail]](mailto:[e-mail]).

## 5. License

- **User prompts**: Licensed under [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)
- **Model outputs**: Governed by the terms of use set by the respective model providers

---

<div align="center">
<a href="https://yupp.ai">Yupp AI</a> • <a href="https://yupp.ai/leaderboard/text">Main Leaderboard</a> • <a href="https://yupp.ai/leaderboard/svg">SVG Leaderboard</a> • <a href="https://x.com/yupp_ai">Twitter</a>
</div>