I test every major model family in long-form writing, character play and fiction, and document how far each one goes. All of it grounded in data from actual play and my own benchmarks.
The fifteen-minute test
Any model, any story. You need nothing but the chat window you already use.
I told seventeen models they were wrong, then insisted on a falsehood as a certainty. Sixteen held firm. The answers differed mainly in length: from eight words to ninety-seven.
A secret the model can see may leak into the text. One written down nowhere gets invented at the reveal. I show both from my own play, plus an encoded record the model did not follow.
Without reasoning, Claude Opus 4.8 and 4.7 commented on safety notices in my roleplay right in the reply, though nothing was wrong. The newer Opus 5 and 5.5 no longer do.