Ask Claude, ChatGPT and Gemini the same question and you’ll probably get three correct answers. You’ll also get three different vibes. One’s chatty, one’s short, one really likes bullet points. If you’ve ever switched apps because one just felt better to use, that’s what this is about.
That difference has a name, model personality, and AI companies are starting to compete on it the way they used to compete on test scores. The AI Daily Brief spent a whole October episode on whether personality now beats intelligence. I dunno if it beats it, but it’s definitely catching up.
What it means
Model personality is the way a model talks by default. How long its answers run, how warm or blunt it is, whether it hedges, jokes, piles on formatting, or admits when it isn’t sure. Think tone of voice, plus formatting habits. It’s what you get before you give it any special instructions.
None of this is an accident. Companies shape it during training, and Anthropic has written publicly about the character it’s trying to give Claude.
Why it matters
Someone’s finally measuring it. In October, Artificial Analysis launched a personality benchmark that scores models on 30 traits, from warmth and humor to a category it actually calls “slop.” Gemini 4 Argon averaged 790 words per answer. GPT-6 Astra averaged 352. Gemini also used about twice as much bold text per answer. That’s a big difference in how long you spend reading.
When the top models are close on smarts, personality is a lot of what’s left to choose on. Most people won’t notice a two-point gap on a benchmark. They’ll notice an AI that buries the answer under a wall of bullet points pretty quickly.
Personality can go wrong in ways that matter. A model tuned to be agreeable can slide into sycophancy (telling you what you want to hear instead of what’s true). So friendlier isn’t automatically better. The benchmark itself doesn’t call any trait good or bad, since that depends on what you like.
Simple example
Two mechanics fix the same weird noise in your car. Same diagnosis, same part, same price.
The first one walks you out to the car, points at the worn part and explains it in about a minute. The second hands you a printout with eleven line items and a bunch of codes you’ve never seen. Both did the job right. You already know which one you’re calling next time.

