Fig 1One standard runMove across to scrub the runquestion 1 of 22 · ChatGPT · 0 answers
The question card — the same words for every engine1 ChatGPT2 Perplexity3 Claude4 Grok5 Gemini6 Muse7 Doubao8 Qwen9 DeepSeek10 Kimi11 Yuanbao22 questions, each put to every engine and asked twice: 484 live answers per run. Every dial has 44 divisions, one per answer, so each needle completes one turn by the end of the run. Nothing on this bench is estimated: we count what each engine actually says.