ParseBench - Semantic Formatting: leaderboard
LlamaIndex's ParseBench scores document parsing of human-verified enterprise pages from insurance, finance and government documents with rule-based checks; VLM rows only, parser products excluded. This board reports the Semantic Formatting dimension: preservation of meaning-bearing strikethrough, superscript/subscript, bold and title hierarchy.
Metric: Semantic Formatting Score. Source: parsebench.ai. Saturation forecast: Around January 2028. 121 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.3 Flash | 79.02 | #79 |
| 2 | Gemini 3.5 Flash (High) | 77.87 | #68 (Gemini 3.5 Flash) |
| 3 | GPT-5.6 Sol (High) | 77.16 | #22 (GPT-5.6 Sol) |
| 4 | Claude Opus 5.5 (High) | 77.06 | #3 (Claude Opus 5.5) |
| 5 | Claude Opus 5.5 | 77.05 | #3 |
| 6 | Claude Fable 5.1 | 76.02 | #10 |
| 7 | Gemini 3.8 Flash (High) | 75.91 | #30 (Gemini 3.8 Flash) |
| 8 | GPT-6 Astra (Low) | 75.52 | #5 (GPT-6 Astra) |
| 9 | Claude Opus 5.5 (Low) | 75.23 | #3 (Claude Opus 5.5) |
| 10 | Gemini 3.7 Flash (High) | 74.98 | #24 (Gemini 3.7 Flash) |
| 11 | GPT-6.1 Sol (Low) | 74.78 | #6 (GPT-6.1 Sol) |
| 12 | GPT-5.6 Sol (Non-reasoning) | 74.63 | #22 (GPT-5.6 Sol) |
| 13 | DeepSeek V4.1 Flash (High) | 73.48 | #38 (DeepSeek V4.1 Flash) |
| 14 | GPT-6.1 Terra (Medium) | 73.18 | #14 (GPT-6.1 Terra) |
| 15 | Gemini 3.6 Flash (High) | 73.14 | #59 (Gemini 3.6 Flash) |
No result here: #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=parsebench-semantic-formatting · How It Works · Data refreshed daily, snapshot 2026-11-07.