List your company

Visual Discrepancy Detection on OddGridBench

91.33Color AccuracyHuman
-1.573222.545946.66570.7841Mar 10, 2026

Evaluation results

MethodLinks
HumanModel Category=Human Performance2026.0391.3369.3382.67789697.3397.6787.47
OddGrid-GRPO2026.0389.564.580.564.590.591.597.582.64
GRPO2026.0388.54467.541.578.5839370.86
OddGrid-GRPOdistance reward=disabled2026.0387.544.56745.580.59191.572.5
OddGrid-GRPOcurriculum-guided scheduling=disabled2026.0387.560.569648488.595.578.43
Qwen3-VL-32BModel Category=Open-source MLLMs2026.038539.552.53976.589.594.568.07
Gemini-2.5-ProModel Category=Proprietary MLLMs2026.0382.59.5266.556.5768849.29
Qwen3-VL-4BModel Category=Open-source MLLMs2026.0373283927.55670.57352.43
Qwen3-VL-8BModel Category=Open-source MLLMs2026.037220.535.51955.56771.548.71
GSPO2026.03705581.55985.585.59575.93
Gemini-2.5-FlashModel Category=Proprietary MLLMs2026.0370721548.554.572.539.79
Qwen3-VL-30B (Moe)Model Category=Open-source MLLMs2026.0362.52120.5254752.56241.5
GPT-5Model Category=Proprietary MLLMs2026.0356.59.52152935.54628.93
Gemini-2.0-FlashModel Category=Proprietary MLLMs2026.0353.5516532.543.547.529
InternVL3.5-38BModel Category=Open-source MLLMs2026.0345.52161029.53749.527.07
SAIL-VL2-8BModel Category=Open-source MLLMs2026.03452.552.521.52430.518.71
LLaVA-OneVision-1.5-8BModel Category=Open-source MLLMs2026.0339.57.594.5213747.523.71
LLaVA-OneVision-1.5-4BModel Category=Open-source MLLMs2026.0327.537.511417.52613.79
Molmo-72BModel Category=Open-source MLLMs2026.0325.5271121217.511
Qwen2.5-VL-72BModel Category=Open-source MLLMs2026.0325152.510.517.517.511.29
Baseline2026.0323512.571922.53117.14
Qwen3-VL-2BModel Category=Open-source MLLMs2026.0323512.571922.53117.14
SAIL-VL2-2BModel Category=Open-source MLLMs2026.0313.52.5327.559.56.14
LLaVA-v1.6-34BModel Category=Open-source MLLMs2026.033.522.52.5332.52.71
RandomModel Category=Baseline2026.0322221.543.52.43
Phi-3.5-visionModel Category=Open-source MLLMs2026.032221.51.51.521.79
Qwen2.5-VL-7BModel Category=Open-source MLLMs2026.0321.50.502.52.552