An intelligence index takes scores from several separate tests, such as reasoning, knowledge, math and coding, and combines them into one number so that models can be ranked at a glance. Independent testing organizations publish such indexes, and they are popular in model announcements and news coverage.
A single number is convenient and also loses information. The result depends on which tests were included, how each was weighted and how they were run. A model can rank high overall and still be weak at the one thing you need. Indexes are revised as tests become saturated or are replaced, so a score from one version is not comparable with a score from another.
An index is also only as good as its parts. If a component test is contaminated or too easy, the total is affected.
Use an index to shortlist models, read the individual component scores, and make the final choice with tests based on your own tasks.