# Model comparison task card
Use with [pick a model by a checked task](/articles/right-model-for-the-job).

Task and source of truth: ______
Eligible accounts, tools and data categories: ______
Required input/output or connection: ______

| Criterion fixed before the run | Candidate A | Candidate B |
| --- | --- | --- |
| Exact model/mode, product and date | | |
| Correct material facts or answers | | |
| Missing/contradictory input handled | | |
| Output usable in required format | | |
| Total time including corrections | | |
| Applicable price/limits | | |
| Privacy and authorization met | | |

Use the same bounded input and prompt for the first comparison. Keep outputs, then inspect against a source or answer key. Test meaningful edge cases. A single run is a screening result, not a general ranking. Recheck after a model, mode, prompt or task changes.
