ResearchBenchmarksCommonsense Reasoning on ARC-C (with Efficiency Metrics)Follow80.6AccuracyWINO+50.720858.477966.23573.9921May 16, 2026Evaluation ResultsMethodMethodLinksAccuracyStepsStep ReductionTPSTPS SpeedupWINO+Backbone=LLaDA-8B-Inst...Backbone=LLaDA-8B-Instruct, Shot=0-shot2026.0580.628.788.9150.618.81WINOBackbone=LLaDA-8B-Inst...Backbone=LLaDA-8B-Instruct, Shot=0-shot2026.0573.8947.415.485.425LLaDABackbone=LLaDA-8B-Inst...Backbone=LLaDA-8B-Instruct, Shot=0-shot2026.0551.87256117.11