FinixDocBench - FinixInner: leaderboard
Part of FinixDocBench (Wang et al., 2026), a financial document parsing benchmark: an internal held-out track of 4,000 camera-captured documents from financial and insurance workflows, namely claim medical records, expense statements, expense receipts, medical examination reports and identity documents. The score from 0 to 100 is the mean over the five document types of the composite parsing score of text, table and reading-order accuracy; higher is better.
Metric: Overall score (0-100): mean over the five document categories of the per-category overall parsing score, the composite of text edit distance, table TEDS and reading-order edit distance; internal held-out track of 4,000 camera-captured documents from financial and insurance workflows: claim medical records, expense statements, expense receipts, medical examination reports and identity documents, 800 each. Source: arxiv.org. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.5 | 76.91 | #256 |
| 2 | Qwen 3 VL 235B A22B Instruct | 76.32 | #522 |
| 3 | Qwen 3.5 397B A17B | 70.83 | #246 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=finixdocbench-finixinner · How It Works · Data refreshed daily, snapshot 2026-11-13.