ResearchBenchmarksModel Selection Ranking on 100-problem mini activation set (held-out)Model Selection Ranking on 100-problem mini activation set (held-out)Follow0.869Pearson rNEX0.590280.662640.7350.80736Feb 5, 2026Evaluation resultsMethodMethodLinksPearson rRegret@1Hit@3NEXSeries=Qwen3-VL-8B, #B...Series=Qwen3-VL-8B, #Bench=52026.020.869204—NEXSeries=Qwen3-4B, #Bench=5Series=Qwen3-4B, #Bench=52026.020.85990—NEXSeries=Qwen3-VL-4B, #B...Series=Qwen3-VL-4B, #Bench=52026.020.782570—NEXSeries=Overall, #Bench=20Series=Overall, #Bench=202026.020.778267—NEXSeries=Qwen3-VL-32B, #...Series=Qwen3-VL-32B, #Bench=52026.020.601205—