🔧 Administration

Manage application configuration and settings

📦 Version livrée

Version exacte servie par ce serveur, à citer telle quelle en cas d'incident.

Commit
Branche
Environnement
Date de build
Origine du SHA

🎯 Benchmark

Test and compare model performance using a specific URL. Enter a URL below and use the "Analyze the test URL" button on each model card.

💡 This URL will be used for all benchmark analyses. Each model will analyze the same content for comparison. Click "Save URL" to add it to your history.

📊 Recent Benchmark Results

Current Pre-Processing Model

Configure which model to use for pre-processing (metadata extraction, language detection). Pre-processing happens before the main analysis.

Configuration actuelle:

Loading...

Current Premium Models

-
Active Models
-
Available Models
Loading current configuration...

Manage Premium Models

Select models to use for premium multi-LLM analysis. Recommended: 4-6 models for diverse perspectives.

Loading available models from OpenRouter...

💡 Understanding Model Metrics

💰 Pricing

Model costs are shown per 1 million tokens (input/prompt pricing).

  • 1 token ≈ 0.75 words in English
  • 20,000 words ≈ 26,666 tokens (a long article or report)
  • Example: $0.30/1M means analyzing a 20k-word article costs ~$0.008

Note: Output tokens (model response) may have different pricing. Check OpenRouter for details.

📄 Context Length

Context window is the maximum text the model can process in one request (input + output).

  • 8K ctx (~24 pages) - Short articles
  • 32K ctx (~96 pages) - Long articles, reports
  • 128K ctx (~384 pages) - Books, extensive documents
  • 200K+ ctx (~600+ pages) - Entire books, encyclopedias

Calculation: 1 token ≈ 0.75 words, ~250 words per page

💡 For Narrative-Metrics:

Most news articles and opinion pieces are 500-5,000 words. Models with 16K-32K context are sufficient for typical analyses. Larger contexts allow the model to understand complex narrative structures across entire documents, improving detection of subtle framing and influence techniques.

🤝 Médias partenaires

Ils font connaître le site et reçoivent une part de ce qu'il gagne. Les nommer est la moindre des transparences : un site qui mesure l'influence doit dire d'où vient la sienne.