ResearchBenchmarksMulti-Task Reasoning on MMLUFollow56.7Pass@1DCRL32.98839.14445.351.456Mar 17, 2026Evaluation ResultsMethodMethodLinksPass@1DCRLBackbone=Qwen3-8B-BaseBackbone=Qwen3-8B-Base2026.0356.7DCRLBackbone=Qwen3-4B-BaseBackbone=Qwen3-4B-Base2026.0352.6DCRLBackbone=Llama3.2-3B-I...Backbone=Llama3.2-3B-Instruct2026.0333.9