Hippocratic AI unveils Polaris 6 Ultra, a healthcare model with bedside intuition

Hippocratic AI just pulled back the curtain on Polaris 6 Ultra, a healthcare language model it says is the first built with bedside intuition — the ability to read a patient's emotional state, think mid-sentence and pick the next line without an audible pause. The company says it stacks six upgrades across how an AI care agent listens, thinks and speaks, including a more realistic voice and what it calls clinical hearing.

The headline stat: Hippocratic ran its 7.4-trillion-parameter safety constellation 1.3 times faster than the previous version, per the company's announcement. That matters because voice AI in healthcare usually forces a trade-off — speech-to-speech models are quick but shallow, while reasoning models are capable but slow. Hippocratic says it refused the choice, rebuilding the whole stack for voice, from GPU numerics down to the breath in the synthesized voice.

Experience is the pitch. CEO and co-founder Munjal Shah says Polaris 6.0 was trained on more than 323 million patient conversations, making it what he calls the only truly experienced frontier model for healthcare. His blunt framing: in healthcare, you don't get credit for reminding a patient, you get credit for the action they take.

Chief science officer and co-founder Subhabrata Mukherjee said full-parameter reinforcement learning at frontier scale gave the model not just safety but common sense, intuition and the warmth of an experienced clinician. Clinician-measured safety has been pinned at its ceiling since Polaris 4, and the company says the minor-harm rate in clinician output testing holds at 0.00%. The improvements patients actually feel — naturalness, perceptiveness, speed — are where the gains landed.

Polaris 6.0 now ships in four tiers so clients can match coverage to workload: Ultra with the full constellation of 31 supervisors for clinical patient-facing conversations, plus Pro, Flash and Nano for sub-clinical and administrative jobs. Hippocratic is also leaning on deeper EHR integration and new flexible pricing, which it says lets the model serve any healthcare client.

The broader scorecard: Hippocratic says its agents have logged more than 250 million patient voice interactions across over 300 live clinical use cases with zero safety issues in production, and it has raised $444 million in total funding from backers including Andreessen Horowitz, General Catalyst, Kleiner Perkins, NVIDIA's NVentures, and Google's CapitalG. Its contracts let the company take responsibility for its agents, and the agents don't diagnose or prescribe.