XL-DocBench (SimpleDoc): leaderboard
XL-DocBench run through the SimpleDoc retrieval agent rather than by the model alone. The strongest configuration on the whole benchmark at 44 percent, and the board rates a scaffold plus a model, not a model.
Metric: Accuracy (%). Source: officeintelligence.github.io. Saturation forecast: Around March 2028. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 44 | #79 |
| 2 | Claude Opus 4.6 | 40.6 | #63 |
| 3 | GPT-5.2 | 38.3 | #110 |
| 4 | DeepSeek V3.2 | 36.7 | #205 |
| 5 | Kimi K2.5 | 34.4 | #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-simpledoc · How It Works · Data refreshed daily, snapshot 2026-10-20.