Compare six criteria
Use the same test task and source material for every model.
- Quality on your real task
- Source and web verification
- File and image support
- Code and tool use
- Speed and cost
- Privacy and administration
Use a fixed rubric
Score correctness, completeness, instruction following, uncertainty handling and usability. Save the answer and settings.
Do not treat model names as permanent
Models, plans and features change. Record the test date and exact version every time.
Use real information only. Ask for missing data and verify important claims.
Create my prompt