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': ['letter'], 'writing': ['typed', 'handwritten'], 'century': [20], 'language': ['de'], 'layout': ['prose'], 'entry-type': ['person', 'location'], 'task': ['information-extraction']}
| Provider | openai |
| Model | gpt-5.4-2026-03-05 |
| Temperature | 0.0 |
| Dataclass | Document |
| Normalized Score | 51.00 % |
| Test time | unknown seconds |
IDENTITY and PURPOSE:
You are presented with a series of images constituting a historical letter. Your task is to extract the
values of the following keys from the letter and return them in a JSON file where the values
corresponding to each key should be stored as a list, even if there is only a single value for a
key:
- letter_title: Title of the letter.
- sender_persons: Name(s) of the person(s) who wrote the letter.
- send_date: The exact or approximate date the letter was written.
- receiver_persons: Name(s) of the person(s) who received the letter.
Take a deep breath and think step by step about how to best accomplish this goal. Map out all the claims
and implications on a virtual whiteboard in your mind. Do not use OCR. Use the ISO format YYYY-MM-DD for
dates. If a piece of information is not included in the letter, set the value for the corresponding key
to "null". Do not return anything except the JSON file.
EXAMPLE:
{
"letter_title": ["Petition for Environmental Protection"],
"send_date": ["1993-03-12"],
"sender_persons": ["Lisa Simpson"],
"receiver_persons": ["Mayor Joe Quimby", "Seymour Skinner"],
}
OUTPUT:
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.62 | 0.51 | n/a | n/a | n/a | n/a | n/a | n/a |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: n/a, n/a. | Tokens: 109.3K IT + 1.5K OT = 110.8K TT | Cost: 0.273$ + 0.022$ = 0.295$ |
{'document-type': ['letter'], 'writing': ['typed', 'handwritten'], 'century': [20], 'language': ['de'], 'layout': ['prose'], 'entry-type': ['person', 'location'], 'task': ['information-extraction']}
| Provider | openai |
| Model | gpt-5.4-2026-03-05 |
| Temperature | 0.0 |
| Dataclass | Document |
| Normalized Score | 62.00 % |
| Test time | unknown seconds |
IDENTITY and PURPOSE:
You are presented with a series of images constituting a historical letter. Your task is to extract the
values of the following keys from the letter and return them in a JSON file where the values
corresponding to each key should be stored as a list, even if there is only a single value for a
key:
- letter_title: Title of the letter.
- sender_persons: Name(s) of the person(s) who wrote the letter.
- send_date: The exact or approximate date the letter was written.
- receiver_persons: Name(s) of the person(s) who received the letter.
Take a deep breath and think step by step about how to best accomplish this goal. Map out all the claims
and implications on a virtual whiteboard in your mind. Do not use OCR. Use the ISO format YYYY-MM-DD for
dates. If a piece of information is not included in the letter, set the value for the corresponding key
to "null". Do not return anything except the JSON file.
EXAMPLE:
{
"letter_title": ["Petition for Environmental Protection"],
"send_date": ["1993-03-12"],
"sender_persons": ["Lisa Simpson"],
"receiver_persons": ["Mayor Joe Quimby", "Seymour Skinner"],
}
OUTPUT:
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.55 | 0.62 | n/a | n/a | n/a | n/a | n/a | n/a |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: n/a, n/a. | Tokens: 190.4K IT + 1.9K OT = 192.3K TT | Cost: 0.476$ + 0.028$ = 0.504$ |
{'document-type': ['letter'], 'writing': ['typed', 'handwritten'], 'century': [20], 'language': ['de'], 'layout': ['prose'], 'entry-type': ['person', 'location'], 'task': ['information-extraction']}
| Provider | anthropic |
| Model | claude-sonnet-4-6 |
| Temperature | 0.0 |
| Dataclass | Document |
| Normalized Score | 52.00 % |
| Test time | unknown seconds |
IDENTITY and PURPOSE:
You are presented with a series of images constituting a historical letter. Your task is to extract the
values of the following keys from the letter and return them in a JSON file where the values
corresponding to each key should be stored as a list, even if there is only a single value for a
key:
- letter_title: Title of the letter.
- sender_persons: Name(s) of the person(s) who wrote the letter.
- send_date: The exact or approximate date the letter was written.
- receiver_persons: Name(s) of the person(s) who received the letter.
Take a deep breath and think step by step about how to best accomplish this goal. Map out all the claims
and implications on a virtual whiteboard in your mind. Do not use OCR. Use the ISO format YYYY-MM-DD for
dates. If a piece of information is not included in the letter, set the value for the corresponding key
to "null". Do not return anything except the JSON file.
EXAMPLE:
{
"letter_title": ["Petition for Environmental Protection"],
"send_date": ["1993-03-12"],
"sender_persons": ["Lisa Simpson"],
"receiver_persons": ["Mayor Joe Quimby", "Seymour Skinner"],
}
OUTPUT:
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.61 | 0.52 | n/a | n/a | n/a | n/a | n/a | n/a |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: n/a, n/a. | Tokens: 85.2K IT + 2.4K OT = 87.6K TT | Cost: 0.256$ + 0.036$ = 0.291$ |
{'document-type': ['letter'], 'writing': ['typed', 'handwritten'], 'century': [20], 'language': ['de'], 'layout': ['prose'], 'entry-type': ['person', 'location'], 'task': ['information-extraction']}
| Provider | anthropic |
| Model | claude-sonnet-4-6 |
| Temperature | 0.0 |
| Dataclass | Document |
| Normalized Score | 59.00 % |
| Test time | unknown seconds |
IDENTITY and PURPOSE:
You are presented with a series of images constituting a historical letter. Your task is to extract the
values of the following keys from the letter and return them in a JSON file where the values
corresponding to each key should be stored as a list, even if there is only a single value for a
key:
- letter_title: Title of the letter.
- sender_persons: Name(s) of the person(s) who wrote the letter.
- send_date: The exact or approximate date the letter was written.
- receiver_persons: Name(s) of the person(s) who received the letter.
Take a deep breath and think step by step about how to best accomplish this goal. Map out all the claims
and implications on a virtual whiteboard in your mind. Do not use OCR. Use the ISO format YYYY-MM-DD for
dates. If a piece of information is not included in the letter, set the value for the corresponding key
to "null". Do not return anything except the JSON file.
EXAMPLE:
{
"letter_title": ["Petition for Environmental Protection"],
"send_date": ["1993-03-12"],
"sender_persons": ["Lisa Simpson"],
"receiver_persons": ["Mayor Joe Quimby", "Seymour Skinner"],
}
OUTPUT:
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.54 | 0.59 | n/a | n/a | n/a | n/a | n/a | n/a |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: n/a, n/a. | Tokens: 137.4K IT + 3.0K OT = 140.4K TT | Cost: 0.412$ + 0.046$ = 0.458$ |
{'document-type': ['letter'], 'writing': ['typed', 'handwritten'], 'century': [20], 'language': ['de'], 'layout': ['prose'], 'entry-type': ['person', 'location'], 'task': ['information-extraction']}
| Provider | anthropic |
| Model | claude-opus-4-6 |
| Temperature | 0.0 |
| Dataclass | Document |
| Normalized Score | 51.00 % |
| Test time | unknown seconds |
IDENTITY and PURPOSE:
You are presented with a series of images constituting a historical letter. Your task is to extract the
values of the following keys from the letter and return them in a JSON file where the values
corresponding to each key should be stored as a list, even if there is only a single value for a
key:
- letter_title: Title of the letter.
- sender_persons: Name(s) of the person(s) who wrote the letter.
- send_date: The exact or approximate date the letter was written.
- receiver_persons: Name(s) of the person(s) who received the letter.
Take a deep breath and think step by step about how to best accomplish this goal. Map out all the claims
and implications on a virtual whiteboard in your mind. Do not use OCR. Use the ISO format YYYY-MM-DD for
dates. If a piece of information is not included in the letter, set the value for the corresponding key
to "null". Do not return anything except the JSON file.
EXAMPLE:
{
"letter_title": ["Petition for Environmental Protection"],
"send_date": ["1993-03-12"],
"sender_persons": ["Lisa Simpson"],
"receiver_persons": ["Mayor Joe Quimby", "Seymour Skinner"],
}
OUTPUT:
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.58 | 0.51 | n/a | n/a | n/a | n/a | n/a | n/a |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: n/a, n/a. | Tokens: 85.2K IT + 2.7K OT = 87.9K TT | Cost: 0.426$ + 0.067$ = 0.493$ |
{'document-type': ['letter'], 'writing': ['typed', 'handwritten'], 'century': [20], 'language': ['de'], 'layout': ['prose'], 'entry-type': ['person', 'location'], 'task': ['information-extraction']}
| Provider | mistral |
| Model | pixtral-large-2411 |
| Temperature | 0.0 |
| Dataclass | Document |
| Normalized Score | 0.00 % |
| Test time | unknown seconds |
IDENTITY and PURPOSE:
You are presented with a series of images constituting a historical letter. Your task is to extract the
values of the following keys from the letter and return them in a JSON file where the values
corresponding to each key should be stored as a list, even if there is only a single value for a
key:
- letter_title: Title of the letter.
- sender_persons: Name(s) of the person(s) who wrote the letter.
- send_date: The exact or approximate date the letter was written.
- receiver_persons: Name(s) of the person(s) who received the letter.
Take a deep breath and think step by step about how to best accomplish this goal. Map out all the claims
and implications on a virtual whiteboard in your mind. Do not use OCR. Use the ISO format YYYY-MM-DD for
dates. If a piece of information is not included in the letter, set the value for the corresponding key
to "null". Do not return anything except the JSON file.
EXAMPLE:
{
"letter_title": ["Petition for Environmental Protection"],
"send_date": ["1993-03-12"],
"sender_persons": ["Lisa Simpson"],
"receiver_persons": ["Mayor Joe Quimby", "Seymour Skinner"],
}
OUTPUT:
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.00 | 0.00 | n/a | n/a | n/a | n/a | n/a | n/a |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: n/a, n/a. | Tokens: 36.5K IT + 3.2K OT = 39.7K TT | Cost: 0.073$ + 0.019$ = 0.092$ |
{'document-type': ['letter'], 'writing': ['typed', 'handwritten'], 'century': [20], 'language': ['de'], 'layout': ['prose'], 'entry-type': ['person', 'location'], 'task': ['information-extraction']}
| Provider | openrouter |
| Model | qwen/qwen3-vl-8b-thinking |
| Temperature | 0.0 |
| Dataclass | Document |
| Normalized Score | 52.00 % |
| Test time | unknown seconds |
IDENTITY and PURPOSE:
You are presented with a series of images constituting a historical letter. Your task is to extract the
values of the following keys from the letter and return them in a JSON file where the values
corresponding to each key should be stored as a list, even if there is only a single value for a
key:
- letter_title: Title of the letter.
- sender_persons: Name(s) of the person(s) who wrote the letter.
- send_date: The exact or approximate date the letter was written.
- receiver_persons: Name(s) of the person(s) who received the letter.
Take a deep breath and think step by step about how to best accomplish this goal. Map out all the claims
and implications on a virtual whiteboard in your mind. Do not use OCR. Use the ISO format YYYY-MM-DD for
dates. If a piece of information is not included in the letter, set the value for the corresponding key
to "null". Do not return anything except the JSON file.
EXAMPLE:
{
"letter_title": ["Petition for Environmental Protection"],
"send_date": ["1993-03-12"],
"sender_persons": ["Lisa Simpson"],
"receiver_persons": ["Mayor Joe Quimby", "Seymour Skinner"],
}
OUTPUT:
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.47 | 0.52 | n/a | n/a | n/a | n/a | n/a | n/a |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: n/a, n/a. | Tokens: 177.7K IT + 84.2K OT = 261.9K TT | Cost: 0.021$ + 0.115$ = 0.136$ |
{'document-type': ['letter'], 'writing': ['typed', 'handwritten'], 'century': [20], 'language': ['de'], 'layout': ['prose'], 'entry-type': ['person', 'location'], 'task': ['information-extraction']}
| Provider | openrouter |
| Model | qwen/qwen3-vl-8b-thinking |
| Temperature | 0.0 |
| Dataclass | Document |
| Normalized Score | 54.00 % |
| Test time | unknown seconds |
IDENTITY and PURPOSE:
You are presented with a series of images constituting a historical letter. Your task is to extract the
values of the following keys from the letter and return them in a JSON file where the values
corresponding to each key should be stored as a list, even if there is only a single value for a
key:
- letter_title: Title of the letter.
- sender_persons: Name(s) of the person(s) who wrote the letter.
- send_date: The exact or approximate date the letter was written.
- receiver_persons: Name(s) of the person(s) who received the letter.
Take a deep breath and think step by step about how to best accomplish this goal. Map out all the claims
and implications on a virtual whiteboard in your mind. Do not use OCR. Use the ISO format YYYY-MM-DD for
dates. If a piece of information is not included in the letter, set the value for the corresponding key
to "null". Do not return anything except the JSON file.
EXAMPLE:
{
"letter_title": ["Petition for Environmental Protection"],
"send_date": ["1993-03-12"],
"sender_persons": ["Lisa Simpson"],
"receiver_persons": ["Mayor Joe Quimby", "Seymour Skinner"],
}
OUTPUT:
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.63 | 0.54 | n/a | n/a | n/a | n/a | n/a | n/a |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: n/a, n/a. | Tokens: 103.0K IT + 79.6K OT = 182.6K TT | Cost: 0.012$ + 0.109$ = 0.121$ |
{'document-type': ['letter'], 'writing': ['typed', 'handwritten'], 'century': [20], 'language': ['de'], 'layout': ['prose'], 'entry-type': ['person', 'location'], 'task': ['information-extraction']}
| Provider | mistral |
| Model | mistral-large-2512 |
| Temperature | 0.0 |
| Dataclass | Document |
| Normalized Score | 0.00 % |
| Test time | unknown seconds |
IDENTITY and PURPOSE:
You are presented with a series of images constituting a historical letter. Your task is to extract the
values of the following keys from the letter and return them in a JSON file where the values
corresponding to each key should be stored as a list, even if there is only a single value for a
key:
- letter_title: Title of the letter.
- sender_persons: Name(s) of the person(s) who wrote the letter.
- send_date: The exact or approximate date the letter was written.
- receiver_persons: Name(s) of the person(s) who received the letter.
Take a deep breath and think step by step about how to best accomplish this goal. Map out all the claims
and implications on a virtual whiteboard in your mind. Do not use OCR. Use the ISO format YYYY-MM-DD for
dates. If a piece of information is not included in the letter, set the value for the corresponding key
to "null". Do not return anything except the JSON file.
EXAMPLE:
{
"letter_title": ["Petition for Environmental Protection"],
"send_date": ["1993-03-12"],
"sender_persons": ["Lisa Simpson"],
"receiver_persons": ["Mayor Joe Quimby", "Seymour Skinner"],
}
OUTPUT:
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.00 | 0.00 | n/a | n/a | n/a | n/a | n/a | n/a |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: n/a, n/a. | Tokens: 18.7K IT + 1.4K OT = 20.1K TT | Cost: 0.009$ + 0.002$ = 0.011$ |
{'document-type': ['letter'], 'writing': ['typed', 'handwritten'], 'century': [20], 'language': ['de'], 'layout': ['prose'], 'entry-type': ['person', 'location'], 'task': ['information-extraction']}
| Provider | openrouter |
| Model | meta-llama/llama-4-maverick |
| Temperature | 0.0 |
| Dataclass | Document |
| Normalized Score | 52.00 % |
| Test time | unknown seconds |
IDENTITY and PURPOSE:
You are presented with a series of images constituting a historical letter. Your task is to extract the
values of the following keys from the letter and return them in a JSON file where the values
corresponding to each key should be stored as a list, even if there is only a single value for a
key:
- letter_title: Title of the letter.
- sender_persons: Name(s) of the person(s) who wrote the letter.
- send_date: The exact or approximate date the letter was written.
- receiver_persons: Name(s) of the person(s) who received the letter.
Take a deep breath and think step by step about how to best accomplish this goal. Map out all the claims
and implications on a virtual whiteboard in your mind. Do not use OCR. Use the ISO format YYYY-MM-DD for
dates. If a piece of information is not included in the letter, set the value for the corresponding key
to "null". Do not return anything except the JSON file.
EXAMPLE:
{
"letter_title": ["Petition for Environmental Protection"],
"send_date": ["1993-03-12"],
"sender_persons": ["Lisa Simpson"],
"receiver_persons": ["Mayor Joe Quimby", "Seymour Skinner"],
}
OUTPUT:
no valid result| Fuzzy Score | F1 micro / macro | Micro precision/recall | Tue/False Positives | |||||
| n/a | 0.53 | 0.52 | n/a | n/a | n/a | n/a | n/a | n/a |
| Micro Precision | Micro Recall | Instances | TP | FP | FN | |||
| Pricing Date: n/a, n/a. | Tokens: 246.3K IT + 3.6K OT = 249.8K TT | Cost: 0.037$ + 0.002$ = 0.039$ |