OCR-MetaReasoning - Layout Semantics: leaderboard
Metric: Answer Score (%). Source: arxiv.org. 18 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 (Medium) | 77.6 |
| 2 | Gemini 3.1 Pro (Preview) (Thinking) | 76.2 |
| 3 | GPT-5.4 Mini (Medium) | 73.6 |
| 4 | Kimi K2.5 (Non-reasoning) | 70.2 |
| 5 | Claude Sonnet 4.6 (Thinking) | 69.9 |
| 6 | Qwen 3 VL 235B A22B Instruct | 69.5 |
| 7 | Qwen 3 VL 235B A22B (Thinking) | 69.2 |
| 8 | GPT-5.4 (Non-reasoning) | 69.1 |
| 9 | Grok 4.20 (Non-reasoning) | 61.9 |
| 10 | GPT-5.4 Mini (Non-reasoning) | 61 |
| 11 | Qwen 3 VL 8B (Thinking) | 60.6 |
| 12 | Qwen 3 VL 30B A3B (Thinking) | 60.5 |
| 13 | GLM-4.6V (Reasoning) | 60.3 |
| 14 | Qwen 3 VL 30B A3B Instruct | 55.1 |
| 15 | Qwen 3 VL 8B Instruct | 49.9 |
Interactive version: theaggregate.ai/benchmark?slug=ocr-metareasoning-layout-semantics · How It Works · Data refreshed daily, snapshot 2026-09-19.