ResearchBenchmarksScience Reasoning on MMLUFollow70AccuracyTLMRE40.56848.20955.8563.491Feb 10, 2026Evaluation ResultsMethodMethodLinksAccuracyTokensAES ScoreTLMREModel Size=7BModel Size=7B2026.02707120.44ATTNPOModel Size=7BModel Size=7B2026.0269.85470.56DS-7BModel Size=7BModel Size=7B2026.0269.71,2450ATTNPOModel Size=1.5BModel Size=1.5B2026.0250.66031.32TLMREModel Size=1.5BModel Size=1.5B2026.0248.41,2280.83DS-1.5BModel Size=1.5BModel Size=1.5B2026.0241.71,8720