A test run is a single execution of a benchmark test using a defined model configuration.
Each run represents how a particular large language model (LLM) — such as GPT-4, Claude-3, or Gemini — performed on a given task at a specific time, with specific settings.
A test run includes:
Together, test runs make it possible to compare models, providers, and configurations across benchmarks in a transparent and reproducible way.
{'document-type': ['book-page'], 'writing': ['printed'], 'century': [20], 'language': ['en', 'de'], 'layout': ['list'], 'entry-type': ['company'], 'task': ['information-extraction']}
| Provider | openai |
| Model | gpt-4o |
| Temperature | 0.5 |
| Dataclass | ListPage |
| Normalized Score | 41.20 % |
| Test time | unknown seconds |
The image you are presented with stems from a digitized book containing lists of companies.
Your task is to extract structured information about each company listed on the page.
About the source:
- The image stems from a trade index of the British Swiss Chamber of Commerce.
- The image can show an alphabetical or a thematic list of companies.
- The companies are mostly located in Switzerland and the UK.
- The image stems from a trade index between 1925 and 1958.
- Most pages have one column but some years have two columns.
- The source itself is in English and German but the company names can be in English, German, French or Italian.
About the entries:
- Each entry describes a single company or person.
- Alphabetical entries have filling dots between the company name and the page number. Dots and page numbers are not part of the data and should be ignored.
- Alphabetical entries seldom to never have locations.
- Thematic entries often have locations.
- Thematic entries are listed under headings that describe the type of business.
- Some thematic headings are only references to other headings, e.g. "X, s. Y".
About the output:
- Answer in valid JSON. The JSON should be an array of objects with the following fields:
- The page ID is given as {page_id}.
- Do not add country information, if it is not directly written with the location.
{
"entry_id": "A unique identifier for the entry, e.g. '{page_id}-1'",
"company_name": "The name of the company or person",
"location": "The location of the company, e.g. 'Zurich' or 'London, UK'. If no location is given, set to null."
]
}
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.43 | 0.41 | 0.44 | 0.42 | 15 | 413 | 534 | 579 |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: 9 months ago, 2025-10-28. | Tokens: 18.8K IT + 8.6K OT = 27.5K TT | Cost: 0.047$ + 0.086$ = 0.134$ |
{'document-type': ['book-page'], 'writing': ['printed'], 'century': [20], 'language': ['en', 'de'], 'layout': ['list'], 'entry-type': ['company'], 'task': ['information-extraction']}
| Provider | openrouter |
| Model | meta-llama/llama-4-maverick |
| Temperature | 0.5 |
| Dataclass | ListPage |
| Normalized Score | 32.40 % |
| Test time | unknown seconds |
- Answer in valid JSON.
- The page ID is given as {page_id}.
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.29 | 0.32 | 0.30 | 0.29 | 15 | 284 | 650 | 708 |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: 9 months ago, 2025-10-28. | Tokens: 27.4K IT + 15.0K OT = 42.4K TT | Cost: 0.004$ + 0.009$ = 0.013$ |
{'document-type': ['book-page'], 'writing': ['printed'], 'century': [20], 'language': ['en', 'de'], 'layout': ['list'], 'entry-type': ['company'], 'task': ['information-extraction']}
| Provider | openai |
| Model | gpt-5 |
| Temperature | 0.5 |
| Dataclass | ListPage |
| Normalized Score | 58.40 % |
| Test time | unknown seconds |
The image you are presented with stems from a digitized book containing lists of companies.
Your task is to extract structured information about each company listed on the page.
About the source:
- The image stems from a trade index of the British Swiss Chamber of Commerce.
- The image can show an alphabetical or a thematic list of companies.
- The companies are mostly located in Switzerland and the UK.
- The image stems from a trade index between 1925 and 1958.
- Most pages have one column but some years have two columns.
- The source itself is in English and German but the company names can be in English, German, French or Italian.
About the entries:
- Each entry describes a single company or person.
- Alphabetical entries have filling dots between the company name and the page number. Dots and page numbers are not part of the data and should be ignored.
- Alphabetical entries seldom to never have locations.
- Thematic entries often have locations.
- Thematic entries are listed under headings that describe the type of business.
- Some thematic headings are only references to other headings, e.g. "X, s. Y".
About the output:
- Answer in valid JSON. The JSON should be an array of objects with the following fields:
- The page ID is given as {page_id}.
- Do not add country information, if it is not directly written with the location.
{
"entry_id": "A unique identifier for the entry, e.g. '{page_id}-1'",
"company_name": "The name of the company or person",
"location": "The location of the company, e.g. 'Zurich' or 'London, UK'. If no location is given, set to null."
]
}
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.59 | 0.58 | 0.59 | 0.59 | 15 | 581 | 406 | 411 |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: 9 months ago, 2025-10-28. | Tokens: 16.7K IT + 69.5K OT = 86.3K TT | Cost: 0.021$ + 0.695$ = 0.716$ |
{'document-type': ['book-page'], 'writing': ['printed'], 'century': [20], 'language': ['en', 'de'], 'layout': ['list'], 'entry-type': ['company'], 'task': ['information-extraction']}
| Provider | openai |
| Model | gpt-5 |
| Temperature | 0.5 |
| Dataclass | ListPage |
| Normalized Score | 41.07 % |
| Test time | unknown seconds |
- Answer in valid JSON.
- The page ID is given as {page_id}.
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.36 | 0.41 | 0.37 | 0.36 | 15 | 358 | 612 | 634 |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: 9 months ago, 2025-10-28. | Tokens: 11.6K IT + 51.1K OT = 62.7K TT | Cost: 0.015$ + 0.511$ = 0.526$ |
{'document-type': ['book-page'], 'writing': ['printed'], 'century': [20], 'language': ['en', 'de'], 'layout': ['list'], 'entry-type': ['company'], 'task': ['information-extraction']}
| Provider | genai |
| Model | gemini-2.5-pro |
| Temperature | 0.5 |
| Dataclass | ListPage |
| Normalized Score | 47.93 % |
| Test time | unknown seconds |
- Answer in valid JSON.
- The page ID is given as {page_id}.
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.48 | 0.48 | 0.47 | 0.49 | 15 | 484 | 543 | 508 |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: 9 months ago, 2025-10-28. | Tokens: 4.2K IT + 13.9K OT = 18.1K TT | Cost: 0.005$ + 0.139$ = 0.144$ |
{'document-type': ['book-page'], 'writing': ['printed'], 'century': [20], 'language': ['en', 'de'], 'layout': ['list'], 'entry-type': ['company'], 'task': ['information-extraction']}
| Provider | openrouter |
| Model | qwen/qwen3-vl-30b-a3b-instruct |
| Temperature | 0.5 |
| Dataclass | ListPage |
| Normalized Score | 25.60 % |
| Test time | unknown seconds |
- Answer in valid JSON.
- The page ID is given as {page_id}.
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.24 | 0.26 | 0.24 | 0.24 | 15 | 242 | 754 | 750 |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: 9 months ago, 2025-10-28. | Tokens: 11.6K IT + 13.5K OT = 25.1K TT | Cost: 0.002$ + 0.009$ = 0.012$ |