Methodology
How to read our comparisons
The basis, assumptions and limits behind DesignCat’s published quality and cost comparisons.
Editorial scores
The current quality scores are DesignCat editorial assessments on a ten-point scale. They express a practical judgment, not a measured probability of success, customer rating or independently validated ranking. Unrated models stay unrated.
Configured costs
Each cost states the relevant settings: output resolution, clip duration, reference inputs or character count. These are calculated credit costs from the published comparison snapshot, not receipts from a recorded run. Check the application’s current estimate before submitting generation.
Normalization and estimates
A hypothetical normalization estimates a common comparison unit that may not be available from the model. It is not an observed output. Speech duration estimates depend on speaking rate; cost per minute can also obscure maximum effect or track duration. Keep these assumptions attached when citing or downloading a comparison.
What is not documented yet
The current snapshot does not include a complete prompt/output corpus, per-run model versions and dates, evaluator protocol, sample counts, failure rates or confidence intervals. We do not infer those from the existing scores.
Recorded studies
A recorded study must identify its task, inputs, references, settings, model version, test date, every evaluated output, scoring method, evaluator and sample count. Test-time costs stay with the study. Updated prices are presented separately so a new price cannot silently change historical evidence.
Use and attribution
Link to the category page and this methodology when discussing a comparison. Downloads preserve the evidence label and normalization assumptions. No statistical confidence or universal model superiority is implied.