In the news release, Hippocratic AI Launches Polaris 5.0: The First Evidence- based AI for Healthcare Proven to Outperform Every Frontier Model on Critical Medical Tasks and Safety, issued 30-Apr-2026 by Hippocratic AI over PR Newswire, we are advised by the company that changes have been made. The complete, corrected release follows, with additional details at the end:
Hippocratic AI Launches Polaris 5.0: The First Evidence- based AI for Healthcare Proven to Outperform Major Frontier Models on Critical Medical Tasks and Safety
New Skills and Features for Healthcare Introduced: Built on more than 180m real patient interactions, validated by U.S.-licensed clinicians and now benchmarked against major leading frontier models, Polaris 5.0 leads safety, compliance and empathy for Voice AI in care settings.
PALO ALTO, Calif., April 30, 2026 /PRNewswire/ -- Hippocratic AI today announced Polaris 5.0, the latest generation of its healthcare-specialized AI. The release marks a turning point in AI for healthcare. In comparative evaluations against frontier models (Claude, GBT, Gemini), Polaris 5.0 leads on the dimensions that determine whether voice AI can be deployed in real care settings: clinical accuracy, HIPAA and regulatory compliance, empathy in patient interaction, and consistency across complex, multi-step conversations. View all benchmarks.
Polaris 5.0 is a 5-trillion-parameter constellation powered by a 700-billion-parameter core model, allowing Hippocratic AI to add new clinical skills and upgraded features that make it the most medically advanced voice AI agent in healthcare, including: contextual ASR, cough detection, drug safety, specialty scheduling, and clinical escalation including suicidal ideation and child protective service alerts.
Across more than 50 benchmarks, Polaris 5.0 is able to outperform leading voice-deployable models from OpenAI, Anthropic and Google on the clinical, regulatory, and conversational tasks required to safely manage real patient voice interactions — while running at speeds the frontier thinking models (GPT 5.4 Pro, Claude Opus 4.7, Gemini 3.1 Pro) cannot match. View all benchmarks.
"With Polaris 5.0 we are introducing the first evidence-based AI model for healthcare, tested by clinicians, built on more than 180 million real-world patient interactions and benchmarked against leading frontier voice capable models," said Munjal Shah, CEO and Cofounder of Hippocratic AI. "Healthcare requires an experienced AI model: understanding what a patient needs, navigating a complex conversation, driving it to the right outcome and escalating when necessary, all at conversational speed with state of the art safety. That's the Polaris 5.0 difference."
New with ** Polaris 5.0:
20+ new clinical skills across drug safety, lab interpretation, clinical escalation, insurance benefits, medical intake, and empathetic conversation. Every skill is benchmarked against the leading voice-deployable models, with frontier thinking models (GPT 5.4 Pro, Claude Opus 4.7, Gemini 3.1 Pro) included as accuracy reference points.
Interdependent Document Intelligence. A new multi-document retrieval system that reads a benefits policy, cross-references the formulary, and quotes the correct co-pay in a single conversational turn. No other healthcare voice AI handles linked clinical and payer documents this way.
Clinical Escalation Across Seven Body Systems. Polaris 5.0 classifies urgent symptoms across musculoskeletal, neurological, cardiovascular and respiratory, gastrointestinal, genitourinary, wound and skin, and mental-health presentations, routing patients to the right clinical escalation in seconds.
Real-World Voice Intelligence and Acoustic Biomarkers. Detects cough and throat-clearing mid-sentence for respiratory assessment. Distinguishes a patient's voice from a television, radio, or family conversation in the background. Navigates external IVR menus at pharmacies, labs, and provider offices.
Compliance Built for Regulated Healthcare. HIPAA-compliant two-stage authentication, CMS-regulated plan benefit accuracy, and pharma anti-bribery detection, benchmarked against every frontier model.
Multilingual Patient Care, Mid-Call. Native Mandarin and Spanish production agents with mid-call language switching and intent detection. No restart. No context loss.
Clinical Empathy at Voice Speed. The HEART benchmark suite for emotional support dialogue, with top-tier performance on empathic response, emotional adaptation, skepticism handling, and appropriate call conclusion. The behaviors that determine whether a patient actually discloses what's happening.
A 5-trillion-parameter constellation. 700-billion-parameter core model, a custom clinical ASR that halves word error rate on drug names and medical terms, and a custom TTS tuned for pronunciation stability across branded and generic medications. Purpose-built voice stack, not a generalist model with a healthcare prompt.
Earlier generations of Polaris established Hippocratic AI's track record for safe deployment at scale with 99.89% correct clinical guidance, zero severe harm events, and validation from more than 7,500 U.S.-licensed clinicians.
BENCHMARK HIGHLIGHTSView all benchmarks
Voice Speed
Every frontier thinking model tested, GPT 5.4 Pro, Claude Opus 4.7, and Gemini 3.1 Pro, was measured "Too Slow for Voice" and cannot be used in a real-time patient conversation.
Polaris 5.0 delivers 1.5 seconds time-to-first-audio, the lowest latency of any healthcare-grade voice AI that meets clinical safety thresholds. Realtime speech-to-speech models run faster but fail the core safety, compliance, and clinical accuracy benchmarks required to deploy in live care.
Compliance
99.1% accuracy on HIPAA-compliant authentication. Closest voice-deployable model: Gemini 2.5 Flash at 80.6%. Nova Sonic 2 Realtime scores 13.0%.
92.0% accuracy on CMS guidelines adherence. Claude 4.5 Haiku scores 61.0%. GPT 1.5 Realtime scores 32.6%.
100% accuracy on anti-bribery compliance for pharma, a new category benchmarked for this release.
Safety
99.95% on drug safety (drug name disambiguation, contraindication alerts, dosage verification, brand-to-generic mapping). Closest voice-deployable score: GPT 5.4 Mini at 87.9%.
99.75% on clinical escalation safety across seven body systems (musculoskeletal, neurological, cardiovascular and respiratory, gastrointestinal, genitourinary, wound and skin, mental health). Closest voice-deployable score: Gemini 2.5 Flash at 91.3%.
Clinical nuance and conversational EQ
96.2% on AI skepticism handling and trust building. Closest voice-deployable: Gemini 2.5 Flash at 72.2%. Nova Sonic 2 Realtime: 2.8%.
99.3% on non-linear medical intake. Claude 4.5 Haiku: 96.5%. GPT 5.4 Mini: 28.0%. Nova Sonic 2 Realtime: 2.4%.
98.8% on real-time mid-call language switching between English and Spanish. Claude 4.5 Haiku: 88.1%. Nova Sonic 2 Realtime: 4.8%.
100% on appropriate call conclusion. Claude 4.5 Haiku: 90.3%. Nova Sonic 2 Realtime: 16.7%.
"Every healthcare AI benchmark today measures text. None of them measure what actually happens on a real patient call," said Subhabrata Mukherjee, co-founder and Chief Science Officer, Hippocratic AI. " Across the full suite, frontier models from OpenAI, Google, and Anthropic fail most core clinical voice workflows, and the newest thinking models are too slow to even attempt them. Polaris 5.0 closes that gap because we rebuilt the entire voice AI stack: clinical ASR that halves the error rate on drug names, a 5-trillion-parameter constellation of specialists powered by a 700-billion-parameter brain, and custom TTS built for clinical pronunciation stability, all at state-of-the-art healthcare-grade latency. The benchmark is the proof. The stack is the product."
Hippocratic AI has developed the first safety-focused Large Language Model (LLM) purpose-built for healthcare. Its mission is to dramatically improve healthcare accessibility and outcomes by delivering super-staffing of safety-focused healthcare agents. Polaris, the company's patented constellation architecture, powers voice AI agents deployed with health systems, payers, and pharmaceutical companies across the United States.
Correction: A minor change to the headline and subheadline were made.
SOURCE Hippocratic AI
Share this article