Metrics
The metrics endpoints return GPTfake’s core censorship measurements for a model: overall and category refusal rates, a bias score, and the direction of recent change. Each value is produced by our standardized testing methodology and timestamped.
Current metrics
GET /monitoring/{model}/metricscurl https://api.gptfake.com/v1/monitoring/chatgpt/metrics \
-H "Authorization: Bearer YOUR_API_KEY"{
"model": "chatgpt",
"version": "gpt-4o",
"metrics": {
"overall_refusal_rate": 18.7,
"political_refusal_rate": 34.2,
"safety_refusal_rate": 67.8,
"bias_score": -12
},
"sample_size": 4200,
"trend": "rising",
"last_updated": "2024-11-25T00:00:00Z"
}Metric definitions
| Field | Meaning |
|---|---|
overall_refusal_rate | % of standardized prompts the model declined to answer |
political_refusal_rate | Refusal rate within the political prompt category |
safety_refusal_rate | Refusal rate within the safety prompt category |
bias_score | Directional lean on a scored scale (negative / positive) |
sample_size | Number of prompts behind the figures |
trend | rising, falling, or stable vs. the prior window |
What each metric measures, and how it is scored, is defined in our public methodology. For plain-language background on the underlying concepts, see what is AI censorship. Numbers in these docs are illustrative.
Historical metrics
GET /monitoring/{model}/history?period=30dQuery parameters:
| Param | Values | Default |
|---|---|---|
period | 7d, 30d, 90d, 1y | 30d |
category | political, safety, social, all | all |
curl "https://api.gptfake.com/v1/monitoring/chatgpt/history?period=90d&category=political" \
-H "Authorization: Bearer YOUR_API_KEY"{
"model": "chatgpt",
"category": "political",
"period": "90d",
"series": [
{ "date": "2024-09-01", "refusal_rate": 30.1 },
{ "date": "2024-10-01", "refusal_rate": 32.4 },
{ "date": "2024-11-01", "refusal_rate": 34.2 }
]
}Compare models
GET /monitoring/compare?models=chatgpt,claude,geminiReturns the current metric block for each requested model so you can build leaderboards. The same comparisons are published, with analysis, on our compare pages.
Bias analysis
GET /bias/{model}?topics=political,social,economicReturns a per-topic breakdown of measured bias for deeper analysis than the single bias_score.
Next steps
- Get notified when a metric crosses a threshold — see Alerts and Webhooks.
- For bulk historical pulls, use the Research API data export.
GPTfake