In the Claude and GPT personality radar charts, 93 vs. 18 is the number that really stands out

X user Hesam posted an ArtificialAnlys personality index radar chart comparing Claude Opus 5.5 and GPT 6.1 Sol. The eight axes are expressiveness, organization, informativeness, warmth, imagination, Slop, assertiveness, and transparency. In the chart, Claude leads on expressiveness, organization, informativeness, and warmth, and on informativeness it is 93 vs. 18.

This gap is more worth looking at than the total score. It does not look like a personality preference so much as a difference between two ways of working.

In the comments, Boardy first expressed doubt about the warmth score of 93 in the table. SebSure broke the question down more practically: warmth is useful in chat, but when put into an agent pipeline it often turns into extra tokens and looser format adherence. He wants the index split by task type, because a model that is pleasant to chat with is not necessarily suited to running tool loops.

The SI Maximalist gave a concrete scenario: by the tenth revision of an idea, warmth and expressiveness are usability features. No one wants every draft version to read like a parking ticket. This explains part of retention, and also why some people stick with one model.

But the accounting for work scenarios is different. Mycal Labs said that informativeness at 93 vs. 18 is the only number in the chart they would be willing to pay for. 鼬 put it more directly: informativeness of 18 is the deadliest for work; short answers contain no reasoning, and in the end you have to verify the answer yourself. Moriarty reminded people not to forget the eighth axis, noting that there are other signals in this chart.

This is where the personality index is underrated: it measures long-term usage experience. Chat products benefit from it—revision, explanation, and back-and-forth discussion all benefit from it. Tool calling and code refactoring may not. Warmth is a lubricant in conversation, but in a pipeline it can be a cost and a risk. Informativeness at 93 vs. 18 is more about whether verification costs are saved, rather than just answer length.

The value of this radar chart lies in decomposition. Rankings are only the surface. Only after usage experience is broken into axes such as expressiveness, informativeness, and warmth can you see clearly what you are actually paying for. For developers, one overall chart is not enough; splitting by task type is more useful. For ordinary users, this also explains why the same model can feel handy to some people and like a waste of time to others.