Fractal AI Model Outperforms GPT-5 on Health Benchmark
AI firm Fractal has launched Vaidya 2.0, a new model focused on healthcare. The company claims it is the first AI to score over 50 on OpenAI's HealthBench (hard), outperforming both GPT-5 and Google's Gemini Pro 3 on the specialized benchmark. The result showcases the growing capability of specialized AI models in vertical industries.
- The model was unveiled at the India AI Impact Summit 2026; Fractal is a key partner in the Indian government's ₹10,300+ crore India AI Mission, which aims to build the nation's sovereign AI capabilities. - OpenAI's HealthBench benchmark is designed to test models on realistic, multi-turn medical conversations rather than simple exam questions. The "hard" subset specifically includes 1,000 challenging scenarios that prior frontier models struggled to solve. - Fractal describes Vaidya 2.0 as a "Healthcare Operating System" built with post-trained reasoning and agentic architectures, enabling it to execute multi-step clinical and administrative workflows, not just answer questions. - Specific real-world applications for the model include "Emergency Assist" for rapid triage guidance, a "Symptom Checker" for public wellness inquiries, and "Patient Journey Assist" for end-to-end support from diagnosis through treatment. - The outperformance of a specialized model highlights a trend in AI development where domain-specific models, trained on curated datasets for industries like healthcare or finance, can achieve higher accuracy and reliability on vertical tasks than larger, general-purpose models. - New York-based Fractal is a long-standing AI firm, founded in 2000, that provides AI and analytics to Fortune 500 companies and holds a valuation of approximately $2.4 billion.
