A purpose-built retrieval-augmented generation system named VITA, designed for low- and middle-income settings, was evaluated against general-purpose large language models on the HealthBench benchmark. The study demonstrates that corpus-specific design can maintain competitive performance even as newer frontier models are released.

  • On 4,023 English-language HealthBench questions, VITA ranked first with 51.9% of possible rubric points, ahead of GPT-5.4 (46.1%), o4-mini (44.3%), Gemini 3.1 Pro (42.6%), and Claude Sonnet 4.6 (37.3%).
  • VITA scored highest on 45.4% of questions when graded by a GPT-4.1 judge.
  • In a robustness test using a neutral open-weight judge, VITA achieved parity with GPT-5.5 and led on points-weighted scores.
  • The system retrieves from a curated corpus including disease-specific guidelines, India-specific antimicrobial resistance data, and resource-limited care protocols.

The results indicate that a purpose-built clinical RAG system remains competitive with frontier LLMs, suggesting that corpus specificity improves grounding despite potentially lower communication polish.