benchgap
Calibration

SimpleVQA → CountBench

CountBench is estimated from SimpleVQA with a linear curve fitted on 8 models measured on both: y = 0.1457·x + 0.8526, R² = 0.34, cross-validated error 3.8 pp. It is used for 0 estimates.

Fitted on

ModelSimpleVQACountBench
Claude Opus 4.565.7%90.6%
Gemini 3 Pro73.2%97.3%
GPT-5.255.8%91.9%
Kimi K2.571.2%94.1%
LFM2.5-VL-3B35.4%87.3%
LFM3-VL-2B36.3%93.0%
Qwen3.5 397B67.1%97.2%
Qwen3.6-27B56.1%97.8%