How to compare AI chat assistants

Compare assistants on five measurements: reasoning depth, usable context, tool use, citation discipline and consistency across repeated tasks. Ignore leaderboard hype.

Compare on the work, not the leaderboard

Five measurements that predict whether an assistant helps you or wastes your time:

1. Reasoning depth. Can it hold a multi-step argument without dropping a premise?
2. Usable context. The advertised window matters less than what it actually remembers at 100k tokens.
3. Tool use. Can it search, run code and read files without you babysitting every call?
4. Citations. When it claims a fact, does it show the source?
5. Consistency. Run the same task ten times. Do you get ten useful answers, or two?

The leaderboard measures what a benchmark rewards. Your workload is the benchmark that matters.