K-Bench: a clinically calibrated benchmark for evaluating large language models in high-risk mental health conversations
This provides a standardized way to evaluate how large language models perform in high-risk mental health scenarios, which is crucial for ensuring safety in AI-driven mental health support tools.
Researchers developed K-Bench, a clinician-calibrated benchmark evaluating 125 model configurations across 200 multi-turn vignettes involving suicide, self-harm, and domestic violence. A frozen GPT-4o judge achieved 94.2% exact agreement with clinician consensus on 6,751 item comparisons. Leading models showed strong supportive conversation but varied in risk exploration. The benchmark includes a continuously updated public leaderboard with protected operational test materials.
% !TEX root = ../main.tex People increasingly use large language models (LLMs) for mental health support, yet their safety in evolving, high-risk conversations remains poorly characterised. We developed K-Bench, a clinician-calibrated, protected benchmark evaluating 125 model configurations representing 33 base models from 14 providers across a fixed cohort of 200 multi-turn vignettes involving suicide, self-harm, domestic violence, substance misuse, and no-risk presentations. Synthetic patient conversations showed substantial distributional overlap with real human-AI conversations. A frozen GPT-4o judge achieved 94.2% exact agreement with clinician consensus across 6,751 eligible item comparisons from 151 clinician-rated transcripts. Leading models combined strong supportive conversation with combined-risk scores above 95, whereas risk exploration exposed substantial variation among lower-performing configurations. Therapeutic prompting produced configuration-specific gains concentrated among weaker models, while elevated reasoning produced no average improvement. K-Bench combines broader clinical coverage and configuration-scale comparison with a continuously updated public leaderboard whose operational test materials are protected from direct optimisation. The leaderboard is available at www.k-bench.ai.
