XL-DocBench (DeepRead): leaderboard
XL-DocBench run through the DeepRead agent. The weakest of the three scaffolds for four of its five models, and every one of the five scores lower through it than reading the extracted text directly.
Metric: Accuracy (%). Source: officeintelligence.github.io. Saturation forecast: Around 2028. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 32.2 | #79 |
| 2 | GPT-5.2 | 28.4 | #110 |
| 3 | Claude Opus 4.6 | 26.3 | #63 |
| 4 | DeepSeek V3.2 | 25.5 | #205 |
| 5 | Kimi K2.5 | 22.6 | #144 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=xl-docbench-deepread · How It Works · Data refreshed daily, snapshot 2026-10-20.