Hippocratic AI has announced the release of Polaris 3.0, a suite of 22 large language models (LLMs) comprising 4.2 trillion parameters, designed to improve patient safety and engagement in non-diagnostic clinical tasks. Developed through real-world patient interactions, the latest version has achieved a clinical accuracy rate of 99.38%, an improvement over previous iterations. The model introduces enhancements in deep-thinking capabilities, clinical documentation, emotional intelligence, multilingual safety, and integration with electronic medical records (EMRs). Specific upgrades include improved speech recognition, new orchestration features for seamless workflow integration, and dialer functions that improve patient connection rates. As a result, patient satisfaction has increased, and the refusal rate for AI interactions has dropped significantly.
Munjal Shah, Co-founder and CEO of Hippocratic AI, emphasized the company’s commitment to safety and effectiveness, highlighting extensive real-world testing involving over 6,000 clinicians and 1.85 million patient calls. The company has also introduced the Real World Evaluation of Large Language Models in Healthcare (RWE-LLM) framework to enhance AI safety across the industry. Shah noted that Hippocratic AI’s focus is on achieving a level of “product perfection” tailored to real-world clinical environments. Polaris 3.0’s development is backed by major investors, including Andreessen Horowitz, NVIDIA, and multiple health systems, positioning the company as a leader in AI-driven healthcare solutions.