article · 2026
Small Models Achieve Large Language Model Performance: Evaluating Reasoning-Enabled AI for Secure Child Welfare Research
Journal of Evidence-based Social Work
School of Social Work bio: Brian E. Perron · Joseph P. Ryan · Zia Qi
Abstract
PurposeThis study develops a systematic benchmarking framework for testing whether language models can accurately identify constructs of interest in child welfare records. The objective is to assess how different model sizes and architectures perform on four validated benchmarks for classifying critical risk factors among child welfare-involved families: domestic violence, firearms, substance-related problems generally, and opioids specifically.Materials and MethodsWe constructed four benchmarks for identifying risk factors in child welfare investigation summaries: domestic violence, substance-related problems, firearms, and opioids (n = 500 each). We evaluated seven model sizes (0.6B-32B parameters) in standard and extended reasoning modes, plus a mixture-of-experts variant. Cohen's kappa measured agreement with gold standard classifications established by human experts.ResultsThe benchmarking revealed a critical finding: bigger models are not better. A small 4B parameter model with extended reasoning proved the most effective, outperforming models up to eight times larger. It consistently achieved "substantial" to "almost perfect" agreement across all four distinct benchmark categories. This model achieved "almost perfect" agreement (kappa = 0.93-0.96) on three benchmarks (substance-related problems, firearms, and opioids) and "substantial" agreement (kappa = 0.74) on the most complex task (domestic violence). Small models with extended reasoning rivaled the largest models while being considerably more resource-efficient.ConclusionSmall reasoning-enabled models achieve accuracy levels historically requiring substantially larger architectures, enabling significant time and computational efficiencies. The benchmarking framework provides a method for evidence-based model selection to balance high accuracy with practical resource constraints before operational deployment in social work research. [Victor, Bryan G.] Wayne State Univ, Sch Social Work, Detroit, MI USA. [Victor, Bryan G.] Wayne State Univ, Merrill Palmer Skillman Inst, Detroit, MI USA. [Stoll, Dragan] ZHAW Zurich Univ Appl Sci, Sch Social Work, Zurich, Switzerland. University; Wayne State University