DocFormBench (MCP-Msoffice): leaderboard
Format accuracy on DocFormBench (2026), a content-aware .docx formatting benchmark of 500 instances, with each model driving the open-source MCP-MSOffice server, which exposes Microsoft Word operations as tool calls, for at most 50 steps: the percent of required formatting attributes 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 MCP-MSOffice baseline (an open-source MCP server that exposes Microsoft Word operations on Windows as structured API calls; default settings, at most 50 inference steps); reasoning (thinking) disabled; higher is better. Source: arxiv.org. Saturation forecast: Around November 2028. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Minimal) | 22.38 | #164 (Gemini 3 Flash) |
| 2 | GLM-4.6 (Non-reasoning) | 18.03 | #486 (GLM-4.6) |
| 3 | GLM-4.6V (Non-reasoning) | 16.38 | #701 (GLM-4.6V) |
| 4 | DeepSeek V3.2 (Non-reasoning) | 15.41 | #406 (DeepSeek V3.2) |
| 5 | Qwen 3 235B A22B 2507 Instruct | 14.44 | #619 |
| 6 | GPT-5 (Minimal) | 12.32 | #167 (GPT-5) |
| 7 | Qwen 3 VL 235B A22B Instruct | 12.11 | #522 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=docformbench-mcp-msoffice · How It Works · Data refreshed daily, snapshot 2026-11-13.