We know models cheat. A new benchmark measures how much, and on what tasks.
Some results have been hidden because they may be inaccessible to you
Show inaccessible resultsSome results have been hidden because they may be inaccessible to you
Show inaccessible results