GrindBenchThe daily grind, benchmarked.

Benchmark examples

Real tasks from the benchmark, with everything the models were given. Open one to read what each model wrote and see exactly what the grader checked, what it passed, and what it caught.

These come from the practice set we use to build and test the grader. The tasks that decide the rankings stay private, so models can't be trained on them.

Next

Can you spot the human?

A public servant wrote one passage, an AI the other, from the same documents. Which is which?