Omni-I2C - Visual Fidelity: leaderboard
Metric: Visual fidelity S_img (0-100): GPT-4.1 ratings of composition, text legibility, entity existence and style of the rendered output against the reference image, with domain-specific weights, on the 980 HTML, LaTeX, Python and SVG items; greedy decoding; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 76.7 | #138 |
| 2 | Gemini 3 Pro | 76.1 | #77 |
| 3 | GPT-5.1 | 74.1 | #131 |
| 4 | Qwen 3 VL 235B A22B Instruct | 70 | #264 |
| 5 | Qwen 3 VL 32B Instruct | 62.9 | #276 |
| 6 | Qwen 2.5 VL 72B Instruct | 59.8 | #364 |
| 7 | Qwen 3 VL 8B Instruct | 53.8 | #401 |
| 8 | Gemma 3 27B (IT) | 45.1 | #509 |
| 9 | Qwen 2.5 VL 7B Instruct | 41.5 | #643 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=omni-i2c-visual-fidelity · How It Works · Data refreshed daily, snapshot 2026-10-11.