Which AI is right?
The benchmark
The first benchmark run hasn't happened yet. When it does, every model's score and every question will be here.
Live: how often each model agreed with the referee
From real panel answers over the last 90 days, by kind of question. Agreement isn't the same as being right, but a model that keeps disagreeing with the others is worth knowing about. We record only which model agreed and how long it took, never what was asked.
Not enough panel answers yet to show this fairly. Each model needs at least 5 answers in a kind of question.
What people thought
100% of 1 ratings were a thumbs up.