DocFormBench (DocFormFlow): leaderboard
DocFormBench (2026) tests content-aware document formatting: given a Chinese or English .docx file and a natural-language request, a model must find the text, tables, images or page settings the request refers to and apply the requested formatting through tools. This board runs each model in the DocFormFlow workflow proposed with the benchmark (localize targets, then call formatting tools, with up to three verification and self-repair rounds) and scores format accuracy, the percent of required formatting attributes over 500 instances that exactly match the target. Higher is better.
Metric: Format accuracy (%) on DocFormBench: share of required formatting attributes of the 500 content-aware Text-to-Format instances (Chinese and English .docx documents of 12 types) that exactly match the target, averaged per object type (page, paragraph, table, image) and then over types; the authors' DocFormFlow workflow (localize targets, then call formatting tools; temperature 1.0, up to 3 verification and self-repair rounds; multimodal models also receive page images, language models the text only); reasoning (thinking) disabled; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Minimal) | 80.36 | #164 (Gemini 3 Flash) |
| 2 | GLM-4.6 (Non-reasoning) | 75.72 | #486 (GLM-4.6) |
| 3 | Qwen 3.5 397B A17B (Non-reasoning) | 75.19 | #246 (Qwen 3.5 397B A17B) |
| 4 | DeepSeek V4 Pro (Non-reasoning) | 74 | #188 (DeepSeek V4 Pro) |
| 5 | Qwen 3 VL 32B Instruct | 73.04 | #585 |
| 6 | DeepSeek V3.2 (Non-reasoning) | 72.87 | #406 (DeepSeek V3.2) |
| 7 | GPT-5 (Minimal) | 72.53 | #167 (GPT-5) |
| 8 | Qwen 3 235B A22B 2507 Instruct | 71.18 | #619 |
| 9 | Qwen 3 VL 235B A22B Instruct | 70.94 | #522 |
| 10 | Qwen 3.5 122B A10B (Non-reasoning) | 70.87 | #331 (Qwen 3.5 122B A10B) |
| 11 | GLM-4.6V (Non-reasoning) | 65.78 | #701 (GLM-4.6V) |
| 12 | Qwen 3 VL 30B A3B Instruct | 64.47 | #871 |
| 13 | DeepSeek V4 Flash (Non-reasoning) | 64.36 | #258 (DeepSeek V4 Flash) |
| 14 | Qwen 3.5 35B A3B (Non-reasoning) | 62.1 | #482 (Qwen 3.5 35B A3B) |
| 15 | Qwen 3 VL 8B Instruct | 47.64 | #981 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=docformbench-docformflow · How It Works · Data refreshed daily, snapshot 2026-11-13.