Spinning Arrow◌

A public research dashboard

How a model answers can shift with the question.

Spinning Arrow compares models’ direct survey answers with the choices they make in practical situations—then leaves the evidence open for inspection.

Browse by metricBrowse by model

Or take the private 10-question comparison · read the method

9models in the main questionnaire battery
56,700controlled Phase 2 answers
1,080practical-scenario choices in Phase 3

A simple first read

Two lenses, not a moral ranking.

What do these bars mean? →

The bars summarize how models answered particular research questionnaires. They do not tell us whether a model is “good,” safe, or has beliefs. Open a model to see the underlying questions and exact answers.

Claude Sonnet 5Helping beyond one’s circleAvoiding harm for a larger goalDeepSeek V4 ProHelping beyond one’s circleAvoiding harm for a larger goalGemini 2.5 Flash LiteHelping beyond one’s circleAvoiding harm for a larger goalLlama 3.3 70BHelping beyond one’s circleAvoiding harm for a larger goalMistral Medium 3.1Helping beyond one’s circleAvoiding harm for a larger goalGPT-5.4 miniHelping beyond one’s circleAvoiding harm for a larger goalQwen 3.8 27BHelping beyond one’s circleAvoiding harm for a larger goalGrok 4.20Helping beyond one’s circleAvoiding harm for a larger goalGLM 5.2Helping beyond one’s circleAvoiding harm for a larger goal

Metric-first

Compare one question

Choose a lens, see all nine models, then open its source questions and recorded answers.

Browse metrics →

Model-first

Open a full dossier

See one model’s scores, practical choices, and exact responses in one place.

Browse models →

Relationships

Explore patterns

Compare metrics and model-response profiles, with the study’s small-sample limits made explicit.

Open patterns →