Show HN: Darts Vision Benchmark(darteval.vercel.app)1 points·by red545·7개월 전·0 commentsdarteval.vercel.appShow HN: Darts Vision Benchmarkhttps://darteval.vercel.app/darts-vision/dashboardDetecting the correct score from a dartboard photo is unexpectedly hard for LLMs.The task appears to stress spatial reasoning: Gemini 3 models lead this benchmark by a decent margin.Counterintuitively, “more reasoning” often reduces accuracy.Even the top-performing model scores only ~36% of darts correctly.0 commentsPost comment—
The task appears to stress spatial reasoning: Gemini 3 models lead this benchmark by a decent margin.
Counterintuitively, “more reasoning” often reduces accuracy.
Even the top-performing model scores only ~36% of darts correctly.