Hippocratic AI Launches Polaris 6 Ultra for Real-Time Autonomous Voice Care

The model features a 7.4-trillion-parameter architecture and is trained on more than 323 million real-world patient conversations to improve its ability to handle complex clinical interactions, patient concerns, and emergency escalations.

Hippocratic AI has launched Polaris 6 Ultra, a next-generation healthcare AI model designed to support autonomous voice interactions through real-time reasoning, clinical speech recognition, and adaptive patient conversations. The model features a 7.4-trillion-parameter architecture and is trained on more than 323 million real-world patient conversations to improve its ability to handle complex clinical interactions, patient concerns, and emergency escalations.

The latest model introduces what the company calls "bedside intuition", a capability that allows the AI to assess patient responses and adjust conversational pacing while speaking. Through mid-sentence reasoning, Polaris 6 Ultra begins planning its next response while the current sentence is still being delivered, reducing conversational delays.

The system reports a time-to-first-sentence latency of 278 milliseconds. Its planning decisions are also logged in a clinician-facing Safety Portal, allowing medical directors to review risks identified or deferred during conversations.

Expanded Architecture and Clinical Speech Recognition

Polaris 6 Ultra expands Hippocratic AI's architecture from 5 trillion to 7.4 trillion parameters. It comprises a 671-billion-parameter primary generative model, supported by 30 specialist supervisor models and 142 continuous production monitors.

The model also introduces Care Voice v3, which generates speech patterns, including pauses, breathing and changes in pacing, based on conversational context. In testing across 629 complex pharmacology names, the company's medical text-to-speech technology recorded 90% pronunciation accuracy, compared with 75% for leading commercial benchmarks.

Another upgrade, Common Sense 2.0, uses full-parameter reinforcement learning across the primary model. Hippocratic AI reported an 18-point improvement on its internal healthcare intelligence benchmark, with performance gains in clinical nurse escalations, warm handoffs and mid-call language switching.

The Clinical Automatic Speech Recognition (CASR v3) system uses patients' longitudinal electronic health records (EHRs) and real-time conversation history to improve speech recognition accuracy.

In a validation cohort of 2,778 clinician-reviewed production calls, misidentified medications and symptoms declined from 1.19 to 0.20 per 100 calls. Errors involving numeric values, patient names and contact records fell from 0.57 to zero. Hallucinated words in noise-only audio dropped from 24% to 0.6%, while background television dialogue incorrectly attributed to patients declined from 64.8% to below 3%.

Munjal Shah, Co-Founder and CEO of Hippocratic AI, said the model was developed using more than 323 million patient conversations, with an emphasis on improving patient actions and outcomes rather than simply delivering reminders.

Stay tuned for more such updates on Digital Health News