What does measured vs estimated vs editorial mean?

Direct Answer

These labels clarify the source of a claim. "Measured" means directly observed through testing. "Estimated" means inferred from public data or pricing. "Editorial" means a qualitative assessment by human review.

In the AI industry, benchmarks are often presented as absolute facts, even when they are highly subjective or based on optimal, vendor-controlled conditions. Labeling data prevents false precision and helps teams understand how much weight to give a specific claim.

The classification system:

• MeasuredValues derived from directly running prompts or tasks across models and recording the outputs. This is empirical evidence.
• EstimatedValues inferred from public documentation, third-party reports, or pricing pages. It is reliable but not directly tested by us.
• EditorialQualitative assessments made by human review (e.g., "best for nuanced writing"). These do not reduce to numbers and require human judgment.

By explicitly stating what we know, what we guess, and what we judge, we aim to build a more trustworthy evaluation surface.

Need help evaluating AI models?

Request an evaluation and we'll look at the tradeoffs for your specific workflow.

This is an early comparison surface, not a lab-grade ranking. Some observations are editorial or estimated rather than directly measured. Use this as one input, not a definitive answer.