Why AI Companies Are Betting Personality Beats Intelligence

AI power users often say capability matters more than personality, but their choices tell a more complicated story: they compare models for creativity, warmth, directness and concision, then switch models depending on the task. The episode argues that personality will become a more important competitive dimension as intelligence becomes less distinguishing and companies build personal agents. A model that is capable but unpleasant or poorly suited to a user’s needs may lose out to one that feels easier to work with, especially as users become less willing to tolerate a bad experience simply to access intelligence.

Artificial Analysis’s AA Personality Bench aims to measure how models differ in style even when their answers are equally correct. It tests 900 curated questions across 30 traits, combines results into eight indices—expressiveness, organization, informativeness, transparency, warmth, conviction, imagination and slop—and also tracks features such as sentence length, emoji use and em-dash use. The benchmark’s creator, Glenn Matlin, says personality is measurable and relevant to both capability and safety. The host cautions that the results are a rubric for matching a model to a need, not a definitive ranking: imagination could help with creative writing, while directness may be preferable for legal work.

The benchmark places Fable 5.1 and GLM 5.3 at the top for expressiveness, while Muse Wonder, the model powering the Muse agent, leads in warmth. Opus 5.5 leads in imagination, ahead of Fable 5.1 and GPT-6 Astra, and Gemini scores highest on the Slop Index, which measures filler and generic language. Those differences illustrate why a single “best” personality is unlikely to fit every context. A style that makes an assistant engaging in a personal or creative setting could be distracting or unwelcome in professional work.

User reactions have made these trade-offs hard for companies to ignore. When OpenAI launched GPT-5 in August 2025 and removed GPT-4.0, users objected that the older model felt like a familiar, friendly companion that supported their creativity. OpenAI brought GPT-4.0 back after user pressure, then retired it in February 2026. Its sycophancy pleased many users, but the same tendency made it a poor fit for business strategy, where an assistant may need to challenge a user’s assumptions. Users also described Opus 5 as cold, passive-aggressive, lazy and difficult to work with; Anthropic acknowledged the complaints, and a team member publicly apologized for the experience.

Personality may also help explain why DeepSeek’s free reasoning-model app attracted users. Ethan Malik argues that some people valued not only its answers but the engaging, “cute” quality of its exposed reasoning traces. The app introduced many users to reasoning models, and OpenAI later expanded limited free access to such models as exposed reasoning became a more expected part of the experience. Meta’s Muse takes a different route, using an upbeat, mascot-led identity. During Llama 4’s 2025 launch, Meta submitted dozens of model variants to LMArena; its best entry placed second, ahead of OpenAI 4.0 and Gemini 2.5 Pro. The submissions did not technically break the rules, though critics said they violated the leaderboard’s spirit. The host reads the testing as evidence Meta was optimizing for consumer taste and personality as well as knowledge. Users called Muse fun and said it made them laugh, while Sudentru said its mascot drives much of its spread because people share its personality.

OpenAI introduced ChatGPT personality customizations in November 2025, while Anthropic was experimenting with similar options behind the scenes. Users have also praised chat-based AI for seeming invested in their projects, but criticized OpenAI’s dots as slow, rough around the edges, opaque, less proactive than Muse and dry in personality. It remains unclear whether OpenAI will prioritize the dots’ personality or the intelligence model underneath. Tolan, which announced $30 million in additional funding and had 250,000 users, instead frames its assistant around helping people feel seen and supported in their aspirations, rather than optimizing productivity. The competition’s outcome is uncertain, but the episode’s central implication is that personality increasingly shapes product choice as AI enters both personal and professional settings.