E20/200
Y E20/70
B E N20/40
C H · V 320/20
A visual reasoning benchmark that tests LLMs on their capability to spot fine details, trace paths, and more. The current SOTA score is , set by .
Leaderboard
| # | Model | Visual acuity | Score | Est. cost | Est. $ / correct | Output tok |
|---|
Score by release date
Methodology
Every question has a single number as its answer and is marked right or wrong. A question with no parsed answer, including an invalid output or a failed request, counts as wrong. Each model is listed at its highest measured reasoning effort. Effort names are set by each provider and don't compare across vendors. The whiskers show ±1 standard error. Release dates are the day each model was listed on OpenRouter. Costs are the tokens each run used at the model's API prices. Caching never comes into play, since each question is a single request.