GrindBenchThe daily grind, benchmarked.
← All findingsFinding 3 of 5

"Best model" depends on the work

97% / 33%
the best any model did on procurement evaluation, and on ATIP redaction

One ranking hides a lot. On procurement evaluation the best model got 97% of the work right. On ATIP redaction the best any model managed was 33%.

  • Procurement evaluation · Claude Opus 5.597%
  • Bilingual equivalence · 4 models97%
  • Performance analysis · Claude Sonnet 5.572%
  • ATIP redaction · Claude Opus 5.533%
The best score any model reached, by kind of work.

So the useful question is never "which AI is best?" It is "which is best at the work I have?", and for some work today the honest answer is none of them.

Find the kinds of work you do →

Next

Same model, different work, different answer

GPT-6 Luna got 54% of PubSec work right and 18% of Banking work. Muse Spark 1.3 scored 64% and 63%. A model's score on one kind of job tells you little about the next.