ParseBench: leaderboard
Enterprise document parsing benchmark over tables, charts, layout, and content faithfulness for multimodal extraction systems.
Metric: Overall Score. Source: parsebench.ai. Status: years away from saturation. 64 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Fable 5.1 | 78.92 |
| 2 | Claude Fable 5 | 70.78 |
| 3 | GLM-5.3 Flash | 70.75 |
| 4 | Gemini 3.1 Pro (Preview) | 69.14 |
| 5 | GPT-5.5 | 67.76 |
| 6 | GPT-5.5 (Non-reasoning) | 64.39 |
| 7 | GPT-5.6 Terra (Non-reasoning) | 64.17 |
| 8 | Claude Opus 4.8 | 63.7 |
| 9 | Claude Opus 4.7 | 63.34 |
| 10 | Gemma 4 31B (IT) | 62.4 |
| 11 | GPT-5.4 (Non-reasoning) | 62.23 |
| 12 | Claude Sonnet 5 | 62.13 |
| 13 | GPT-5.6 Sol (Non-reasoning) | 62.12 |
| 14 | Qwen 3 VL 8B Instruct | 61.97 |
| 15 | Qwen 3.8 27B | 61.21 |
Interactive version: theaggregate.ai/benchmark?slug=parsebench · How It Works · Data refreshed daily, snapshot 2026-09-05.