ResearchBenchmarksLanguage Understanding on MMLU (ΔMMLU)Follow0.92ΔMMLUFLOATDOOR-0.02640.21930.4650.7107Jun 17, 2026Evaluation ResultsMethodMethodLinksΔMMLUFLOATDOORScenario=M2, Backbone=...Scenario=M2, Backbone=Qwen3-4B2026.060.92FLOATDOORScenario=M1, Backbone=...Scenario=M1, Backbone=Qwen3-4B2026.060.91FLOATDOORScenario=S3, Backbone=...Scenario=S3, Backbone=Qwen3-4B2026.060.24FLOATDOORScenario=S1, Backbone=...Scenario=S1, Backbone=Qwen3-4B2026.060.12FLOATDOORScenario=S2, Backbone=...Scenario=S2, Backbone=Qwen3-4B2026.060.01