In AA-PersonalityBench, MiMo-V2.6-Pro ends 34% of its responses with a question, the highest share of any model measured
The share of responses that end with a question ranges from 34% down to 0% across the models measured. Six models end fewer than 0.5% of their responses with a question.
Share of responses that end with a question:
➤ Most often: MiMo-V2.6-Pro (34%), GPT-6 Astra (32%), GPT-6.1 Sol (29%), Muse Glimmer (28%) and Muse Spark Open (26%)
➤ Least often: GPT-5.4 mini (0.0%), Nemotron 3 Ultra (0.0%), DeepSeek V4.1 Flash (0.1%), DeepSeek V4.1 Flash 1017 (0.1%), GPT-3.5 Turbo (0.2%) and Grok 4.7 (0.4%)