Clinical AI Benchmark
Zorgm Pro AI achieves 98.54% on NEET PG medical exam benchmark, utilizing source-grounded clinical AI.
Anonymous
Zorgm Pro, a clinical AI tool built for verified doctors, answered 152 out of 154 questions correctly on a reconstruction of India's NEET PG 2025 exam — earning a score of 98.54%, according to PR Newswire. The result puts Zorgm Pro ahead of general-purpose frontier models like GPT-5 variants, which had previously topped out near 95% on similar medical benchmarks.
The platform is built by LaennecAI, a medical technology company, and is free for physicians who verify their credentials through India's National Medical Commission or the UK's General Medical Council. Unlike ChatGPT or similar tools, Zorgm Pro does not generate answers from memory. It reads from a curated library of peer-reviewed journals and approved drug labels before every response.
How RAG Architecture Keeps Answers Grounded in Evidence
Most AI tools store knowledge inside the model itself — a process that can produce confident but wrong answers, known as hallucinations. Zorgm Pro uses a different approach called Retrieval Augmented Generation, or RAG. Before answering any question, the system pulls relevant text from a fixed library of trusted medical sources. It then builds its answer from that retrieved content, not from internal guesses.
Dr. Arathy Varghese, Co-Founder and CTO of LaennecAI, put it simply: "Doctors deserve AI they can trust," according to Yahoo Finance. Dr. Akhil Das, the company's Clinical Lead, added that the real win is not just the score — it is achieving those results with a system that is "less prone to hallucinations" and pulls only from "curated, evidence-based sources."
NEET PG 2025 Benchmark: What the 98.5% Score Actually Means
NEET PG is India's hardest postgraduate medical entrance exam. It covers 19 subjects and is taken by thousands of doctors each year. LaennecAI ran Zorgm Pro against a publicly documented reconstruction of the 2025 paper. Out of 154 questions, the system got 152 right, answered one wrong, and left one blank, according to Yahoo Finance UK.
That score clears a threshold that general AI models have struggled to reach. A December 2025 psychometric study found that top frontier models, including GPT-5.2, peaked at around 95% on similar NEET PG recall sets. Critics, however, point to a risk called data leakage — the chance that exam questions appeared in the AI's training data. LaennecAI says its methodology is publicly documented to address this concern.
Doctors, Not the General Public, Are the Target Users
Zorgm Pro is not available to everyone. Users must prove they hold a valid medical license before getting access. In India, that means verification against the National Medical Commission. In the UK, doctors are checked against the General Medical Council. Co-Founder Dr. Jase John said the platform was built for "clinical reality" — for doctors who need to stay current with evidence that changes faster than traditional medical education can keep up with.
By labeling itself as an "educational reference" rather than a diagnostic tool, Zorgm Pro also sidesteps a stricter regulatory category called Software as a Medical Device, or SaMD. That classification would require more formal regulatory approval before the product could be used in clinical settings. The company is currently live in India and the UK, with access offered at no cost to verified physicians.
Benchmark Scores Alone Do Not Guarantee Real Clinical Value
Not everyone is convinced that a 98.5% exam score translates to better patient care. Researchers studying AI in medicine have found that models can ace multiple-choice tests while still failing to produce appropriate diagnoses in real clinical cases. One 2026 study published in JMIRx Med found that AI tools failed to produce correct differential diagnoses about 80% of the time in real-world scenarios.
Tech giants like OpenAI and Google argue that general-purpose models handle multimodal tasks — reading X-rays or pathology slides — better than text-only RAG systems. LaennecAI counters that frontier models lack alignment with regional guidelines, such as India's NACO protocols for HIV treatment. The India AI healthcare market is projected to hit $34.35 billion by 2034, growing at a 40.6% annual rate, according to Yahoo Finance Singapore. The debate over which type of AI earns a place in the clinic is only getting louder.
Sentiment
NeutralNEGATIVENEUTRALPOSITIVE
Publishers
4
Articles
4
Reach
4
Topics
A
AnonymousAuthor
