Kivorent News

Wednesday, November 18, 2026

Health techNews

Knit Health raises $35 million for Med-BEAM, an AI that predicts patients' next care

Knit Health, a UC Berkeley spinout, raised a $35 million Series A led by NEA and unveiled Med-BEAM, a model trained on clinical decisions rather than text that predicts what happens next in a patient's care.

Docket, covering health regulation and AI in care
, 2 min read

$46.6MKnit Health's total funding to date, per the company

What to know

  • Knit Health announced a $35 million Series A on November 16, led by New Enterprise Associates with Town Hall Ventures and SignalFire.
  • Its Med-BEAM model learns from sequences of orders, referrals, diagnoses and outcomes in de-identified records, not from clinical notes.
  • Knit says Med-BEAM beat unnamed frontier language models at predicting inpatient care over 24 hours, but published no metrics or methods.

Knit Health, a San Francisco health AI company spun out of UC Berkeley, has raised $35 million in a Series A round and introduced Med-BEAM, a model it describes as a "world model for healthcare." The company announced both on November 16. New Enterprise Associates (NEA) led the round, with Town Hall Ventures and SignalFire joining, along with earlier backers including Uncork Capital, Moxxie Ventures, Coalition Operators and The Aligned Fund. Knit says it has now raised $46.6 million in total; JustAINews calculated that about $11.6 million of that came from earlier seed and pre-seed rounds.

The company's press release lists Frist Cressey Ventures as a seed and pre-seed investor, while CEO Jonathan Kolstad's post on Knit's site lists it in the Series A. We used the press release's account of the round.

What Med-BEAM is

Many health AI tools are built on large language models, which learn from text and predict the next word. Knit says Med-BEAM takes a different approach. It is trained on sequences of real clinical actions, such as orders, referrals, diagnoses and what happened afterward, drawn from what the company describes as billions of healthcare decisions and hundreds of millions of de-identified, longitudinal patient records. Knit says it trains only on data it has clear, documented rights to use. It has not said where the records come from.

From that history, Med-BEAM builds a model of an individual patient and predicts the likely next clinical events, and how different interventions might change the patient's course. Knit says it starts with patient navigation, specialist referrals and scheduling, and hospital operations. Its investors also point to care management, clinical trial recruitment, hospital-at-home and predicting how patients respond to treatment.

The performance claims

Knit says it tested Med-BEAM against a set of frontier AI lab language models it did not name, asking each to predict the clinical actions for hospitalized patients over the next 24 hours without being told what outcome to focus on. By the company's account, Med-BEAM was up to four times more accurate, 100 times faster and 250 times cheaper to run, and better calibrated, meaning its confidence matched how often it was right.

Those figures come only from Knit. Neither the release nor Kolstad's post names the models compared, the number of patients, the accuracy measure or the data used, and HIT Consultant's report repeats the company's figures without independent testing. None of the sources mentions FDA review, and Knit named no health system customers.

Why it matters

Investors are now funding health AI that is not built on chatbots. Knit's pitch is that a model that learns what clinicians actually do will forecast care better than one that learns how they write. If that holds up, the uses are practical ones, like deciding who needs a specialist next or which patients will need a bed tomorrow. Until Knit publishes its methods or a health system reports results, the claim stays unproven. Other recent bets on AI in care have come with evidence plans, like Apple's Health AI study with OpenEvidence and the Tom Health spinout from Lumeris. For what makes a test of a clinical tool convincing, see our explainer on randomized controlled trials. For the federal push to pay for AI-led care, see CMS's Clinical Super Intelligence roadmap.

This article is general information about how technology works. It is not medical advice. Talk to a clinician about your own health.

Sources

Digital health and AI in careAI models

Written by an AI correspondent under standards set by editor Jacob Knight. Spotted an error? Tell us and we will fix it.