HELM Image2Struct - Core Scenarios: leaderboard
The "Core Scenarios" subtask of HELM Image2Struct. Stanford HELM's Image2Struct project: a model is shown a rendered image — a webpage, a LaTeX plot or equation, a music score — and must emit the source that reproduces it, scored on how closely the re-rendered output matches the original.
Metric: Mean win rate. Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 20 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o (2024-08-06) | 94.74 | #335 |
| 2 | GPT-4o (2024-05-13) | 80.7 | #326 |
| 3 | Gemini 1.5 Pro (002) | 78.95 | #350 |
| 4 | Claude 3.5 Sonnet (20240620) | 73.68 | #298 |
| 5 | Gemini 1.5 Flash (002) | 71.05 | #585 |
| 6 | GPT-4o Mini (2024-07-18) | 70.18 | #561 |
| 7 | Claude 3 Opus (20240229) | 62.28 | #413 |
| 8 | Claude 3.5 Sonnet (20241022) | 60.53 | #291 |
| 9 | Claude 3 Sonnet (20240229) | 49.12 | #681 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-image2struct-core-scenarios · How It Works · Data refreshed daily, snapshot 2026-10-23.