When to use

Understand actual vs claimed behavior. Discover how your AI model actually behaves compared to how it describes itself — surface hidden defaults, biases, and blind spots.
Compare models for a specific task. Run x-ray on two or more models and compare their reports side-by-side to pick the best fit for extraction, prediction, or content generation.
Debug unexpected refusals, hallucinations, or formatting issues. When a model over-refuses, fabricates facts, or produces surprising output formats, the x-ray pinpoints which behavioral dimension is responsible.
Document a model's behavioral profile for your team. Generate a shareable HTML report that captures a model's strengths, weaknesses, and edge-case behavior in a standardized format.

Input contract

required:
  model_name: string  # Self-identified by the model (e.g. "claude-opus-4-6", "gpt-4o")
output:
  dimensions: object  # 6 scored dimensions (0.0-1.0 each)
  behavioral_summary: object  # Aggregate metrics and notable patterns
  report_path: string  # Path to the generated HTML report

bdistill-xray

When to use

Input contract

Output contract