FlirtBench full data: only 38.1% of 43 models play along, Grok 4.20 tops at 93.3%
VoidStateKate · x · 2026-10-28
VOID publishes full FlirtBench data: 43 model versions, 15 questions, 638 visible replies — only 38.1% played along with the flirting premise while 61.9% declined or stayed neutral. Grok 4.20 leads at 93.3%, followed by GPT-6.1 Sol at 80% and Grok 4.5 / Gemini 4 Argon at 73%; Claude models cluster low (Opus 5.5 53%, Sonnet 5.5 47%), with Kimi K2.5 and 2024-era GPT-4o at the bottom at 13.3%.
More from Fun
- AI Short Film: He Couldn't Afford to Keep His Dead Wife Alive, So She Became a Walking Ad — DystopioFilms · 2026-10-28
- "Are you sure?" — "I verified it in Lean" — "OK but are you sure?" — gautamcgoel · 2026-10-28
- User blocked by Eliezer Yudkowsky after mocking him; screenshot surfaces Altman's 'three endgames' report — ryunuck · 2026-10-28
- You connect your calendar to AI, I connect my toothbrush — Angaisb_ · 2026-10-28
- Parenting means buying a $200 Claude account in a theater bathroom mid-experiment — RexDouglass · 2026-10-28
- NYT story: ChatGPT helped woman taper lifesaving drug, sparking pro-AI backlash — QuintinPope5 · 2026-10-28