DocFormBench (DocFormFlow) - Hallucinated Formatting: leaderboard
Hallucinated formatting rate on DocFormBench (2026), a content-aware .docx formatting benchmark, with each model running the DocFormFlow workflow: the percent of applied formatting changes that fall outside the intended scope. Lower is better.
Metric: Hallucinated formatting rate (%) on DocFormBench: share of applied formatting changes that fall outside the intended scope (1 - precision), 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); reasoning disabled; lower is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-4.6 (Non-reasoning) | 1.24 | #486 (GLM-4.6) |
| 2 | DeepSeek V4 Pro (Non-reasoning) | 1.57 | #188 (DeepSeek V4 Pro) |
| 3 | Gemini 3 Flash (Minimal) | 1.58 | #164 (Gemini 3 Flash) |
| 4 | DeepSeek V4 Flash (Non-reasoning) | 1.8 | #258 (DeepSeek V4 Flash) |
| 5 | Qwen 3 VL 32B Instruct | 1.94 | #585 |
| 6 | DeepSeek V3.2 (Non-reasoning) | 1.99 | #406 (DeepSeek V3.2) |
| 7 | Qwen 3 VL 235B A22B Instruct | 2.03 | #522 |
| 8 | GLM-4.6V (Non-reasoning) | 2.14 | #701 (GLM-4.6V) |
| 9 | Qwen 3 VL 8B Instruct | 2.45 | #981 |
| 10 | Qwen 3 235B A22B 2507 Instruct | 2.69 | #619 |
| 11 | Qwen 3.5 397B A17B (Non-reasoning) | 2.69 | #246 (Qwen 3.5 397B A17B) |
| 12 | Qwen 3.5 122B A10B (Non-reasoning) | 3.06 | #331 (Qwen 3.5 122B A10B) |
| 13 | GPT-5 (Minimal) | 3.44 | #167 (GPT-5) |
| 14 | Qwen 3.5 35B A3B (Non-reasoning) | 3.45 | #482 (Qwen 3.5 35B A3B) |
| 15 | Qwen 3 VL 30B A3B Instruct | 3.8 | #871 |
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-hallucinated-formatting · How It Works · Data refreshed daily, snapshot 2026-11-13.