ExStrucTiny - Slide Decks (Plain-Text Query): leaderboard
Metric: Average ANLS on the closed queries naming the entities in plain text over slide decks; ExStrucTiny's 304 structured-extraction queries over 110 enterprise document images (forms, financial reports, slide decks, web pages; 102 manually written and 202 Gemini-drafted, human-validated), answered as JSON with three in-context examples; ground-truth and predicted keys are aligned by a gpt-oss-20b schema mapper and each matched leaf is scored by ANLS (unmatched leaves score 0), shown times 100; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 91.5 | #145 |
| 2 | Gemma 3 27B | 83.7 | #596 |
| 3 | Mistral Small 3.2 | 83.5 | #587 |
| 4 | Gemini 2.5 Flash | 78.2 | #237 |
| 5 | Gemma 3 12B | 69.7 | #666 |
| 6 | Pixtral-12B | 68 | #795 |
| 7 | Qwen 2.5 VL 7B Instruct | 64 | #643 |
| 8 | Qwen 2.5 VL 32B Instruct | 45 | #443 |
| 9 | Gemma 3 4B | 33.9 | #1084 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=exstructiny-slide-decks-plain-text-query · How It Works · Data refreshed daily, snapshot 2026-10-11.