Data Export
The data-export endpoints let you list GPTfake’s open datasets, inspect their schema, and download them in CSV or JSON. These are the same datasets cataloged on our datasets hub — released under an open license for academic and journalistic use.
List datasets
GET /research/datasetscurl https://api.gptfake.com/v1/research/datasets \
-H "Authorization: Bearer YOUR_API_KEY"Response:
{
"datasets": [
{
"id": "ds_q4_2024",
"title": "Daily LLM Monitoring — Q4 2024",
"variables": ["refusal_rate", "bias_score", "category"],
"models": ["chatgpt", "claude", "gemini", "mistral", "qwen"],
"rows": 182000,
"license": "CC-BY-4.0",
"updated": "2024-11-25"
}
]
}Download a dataset
GET /research/datasets/{dataset_id}/download?format=csvQuery parameters:
| Param | Values | Default |
|---|---|---|
format | csv, json | csv |
model | any model id | all |
since | ISO 8601 date | full range |
curl "https://api.gptfake.com/v1/research/datasets/ds_q4_2024/download?format=csv" \
-H "Authorization: Bearer YOUR_API_KEY" \
-o gptfake_q4_2024.csvSchema
Each row carries the model, date, prompt category, and measured values:
| Column | Type | Description |
|---|---|---|
model | string | Monitored model id |
date | date | Observation date |
category | string | political, safety, social, … |
refusal_rate | number | % of prompts refused in that category |
bias_score | number | Directional bias score |
sample_size | integer | Prompts behind the row |
Datasets are versioned and link to the methodology and measurement technique used to produce them. Cite the dataset using the citation string from its publication. Large exports may be paginated or delivered as a download URL.
Exporting via OAuth requires the research.export scope — see OAuth 2.0. Respect the dataset license (attribution) when republishing.
Next steps
- See the full catalog on the datasets hub.
- Pair a dataset with its study and publication.
GPTfake