Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Loading page…
Preferred languages
More languages supported by Google Translate
Open the selector to load the current language catalog.
Translations are provided by Google Translate.
Multimodal & Grounded benchmark
ParseBench-2: severity-aware document parsing (LlamaIndex leaderboard). Every model the catalog carries a published ParseBench-2 value for, ranked by that value.
Scores document parsers and vision-language models on 4,317 pages from 2,645 documents in six dimensions: text, formatting, tables, charts, visual grounding and forms, with severity-aware scoring and the listed cost per page. A separate release from ParseBench-1; the scores are not comparable. 21 models in the ModelScale catalog have a published value.
Each point is a model with a published ParseBench-2 value and complete token prices, on the 75% input / 25% output blend. Frontier models are spaced evenly along the cost axis.
21 models with a published ParseBench-2 value, ordered by that value, highest first. Models the source has not scored on this benchmark are not listed — they are unmeasured, not last.
| Rank | Model | Provider | Severity-capped page scores, averaged per dataset with fixed weights; six dimension scores and their unweighted mean, in percent |
|---|---|---|---|
| 1 | Anthropic | 87.11 | |
| 2 | 85.32 | ||
| 3 | OpenAI |
What the benchmark measures and how its value is published, from the source’s own record.
ParseBench-2 (ParseBench-2: severity-aware document parsing (LlamaIndex leaderboard)) is a multimodal & grounded benchmark published by ParseBench-2: Severity-Aware Evaluation of Document Parsing. Scores document parsers and vision-language models on 4,317 pages from 2,645 documents in six dimensions: text, formatting, tables, charts, visual grounding and forms, with severity-aware scoring and the listed cost per page. A separate release from ParseBench-1; the scores are not comparable. Its evaluation set is described by the source as 4,317 page-level test cases from 2,645 documents in 13 datasets. The source rates its difficulty as charts and visual grounding separate the field; several systems that lead on text and tables score near zero on charts.
Each model's result is published by ParseBench-2: Severity-Aware Evaluation of Document Parsing as severity-capped page scores, averaged per dataset with fixed weights; six dimension scores and their unweighted mean, in percent. ModelScale prints that value to 2 decimal places, exactly as published, and never converts or rescales it. This page ranks the 21 models that carry a published value, highest value first; a model the source never scored is left out rather than ranked at zero. The source does not state whether a higher value is the better result, so neither does this page.
All leaderboards · Multimodal & Grounded capability leaderboard
| 4 | 81.67 |
| 5 | OpenAI | 78.25 |
| 6 | 77.73 |
| 7 | 76.70 |
| 8 | 76.67 |
| 9 | 75.81 |
| 10 | 75.80 |
| 11 | Meta | 75.66 |
| 12 | OpenAI | 75.36 |
| 13 | Anthropic | 75.07 |
| 14 | OpenAI | 74.59 |
| 15 | Alibaba | 74.48 |
| 16 | Anthropic | 71.24 |
| 17 | OpenAI | 70.21 |
| 18 | 63.64 |
| 19 | 62.18 |
| 20 | Alibaba | 42.92 |
| 21 | 42.76 |
Evidence key: Observed
Rows are ordered by the value ParseBench-2: Severity-Aware Evaluation of Document Parsing published, highest first. The source does not state whether a higher value is the better result, so this page does not either: for a benchmark that measures a rate of failure, read the table from the bottom.