Funding & VC

Knit Health Raises $35M Series A to Scale Its Med-BEAM Clinical Model

Knit Health raised a $35M Series A led by NEA to deploy Med-BEAM, which scored 0.565 average precision predicting clinical actions versus 0.228 for Claude Opus 4.8.

By Amara Osei

3 min read

Knit Health Raises $35M Series A to Expand Med-BEAM Deployment
Knit Health Raises $35M Series A to Expand Med-BEAM DeploymentAI-generated

What's News

  • Knit Health raised a $35 million Series A led by NEA, announced November 16, 2026.
  • Med-BEAM scored 0.565 average precision predicting next clinical actions, versus 0.228 for Claude Opus 4.8.
  • The benchmark used records for 200 emergency-department admissions excluded from Med-BEAM's training data.
  • The Series A follows an $11.6 million seed co-led by Uncork Capital and Frist Cressey Ventures.
  • Knit's earlier model used Truveta data covering more than 130 million patients.

Knit Health has raised a $35 million Series A led by NEA to expand deployments of Med-BEAM, its healthcare world model, the company announced on November 16, 2026. Frist Cressey Ventures, Town Hall Ventures and SignalFire joined the round, alongside continued backing from Uncork, Moxxie, the Aligned Fund and Coalition Operators.

The fresh capital will grow Knit's scientific and engineering teams and put Med-BEAM into more clinical settings. The model is trained on billions of healthcare decisions across hundreds of millions of patient records, learning how conditions evolve, what clinicians do and what follows different interventions, according to the company.

What does Med-BEAM actually do?

Knit's first applications sit in patient navigation, specialist routing, scheduling and hospital operations. Med-BEAM supplies clinical context while language models act as interfaces for patients, clinicians and AI agents — a deliberate division of labor.

The company is aiming squarely at the machinery that connects patients to existing care. Referrals lean on professional relationships. Scheduling depends on rigid rules and appointment availability. Clinical pathways run on simplified categories and scores. The result, Knit argues, is patients waiting for suitable care while clinicians handle work others could do better.

Its example is blunt: a surgeon whose clinic fills with patients who don't need surgery, while patients who would benefit from an operation wait weeks or months.

Knit doesn't want to own every application. It plans to provide a shared clinical-intelligence foundation that health systems, technology companies and developers can build on, rather than each rebuilding the underlying intelligence. That gives Knit a platform position beyond its initial navigation and operations products.

How did Med-BEAM perform against Claude?

Benchmark documentation published November 16, 2026 tested Med-BEAM against Anthropic's Claude Opus 4.8, Claude Sonnet 5 and Claude Haiku 4.5. The task drew on records for 200 patients admitted to the hospital from the emergency department — none of them used in Med-BEAM's training.

Given each patient's history through admission and a full list of possible actions, the models predicted the next 24 hours of care across medications, procedures, lab orders, imaging, nursing actions and referrals. No target outcome was chosen in advance.

The numbers favor Knit heavily. Med-BEAM scored 0.565 average precision on a 0-to-1 measure of how well each model ranked the care that actually occurred. Opus 4.8 scored 0.228, Sonnet 5 scored 0.222 and Haiku 4.5 scored 0.124 — a 2.48x to 4.56x advantage for Med-BEAM on this task.

Speed and cost comparisons used different methods, and Knit discloses that openly. Measured language-model response times ranged from 216 seconds for Haiku 4.5 to 408 seconds for Opus 4.8, with Sonnet 5 at 307 seconds. Med-BEAM's roughly two-second figure was an estimate. Knit put per-patient costs at 37 to 257 times lower than the language models, using actual Anthropic benchmark usage at published rates for the Claude models, and for Med-BEAM, inference time and hourly compute cost multiplied by three to match the provider margin.

What are the benchmark's limits?

Knit itself calls the evaluation an initial result on one defined prediction task. Broader benchmarks across additional tasks, models and clinical settings are planned. The distinction matters: predicting subsequent clinical actions is an early test, while the company's causal-AI objective concerns how alternative decisions would change patient outcomes — a harder claim this benchmark doesn't address.

The financing follows an $11.6 million seed round co-led by Uncork Capital and Frist Cressey Ventures, according to a company-hosted account dated May 12, 2026. Moxxie Ventures led the pre-seed with participation from Coalition Operators. Knit is a UC Berkeley spin-out, and its earlier model used Truveta data covering more than 130 million patients to learn real clinical-practice patterns.

The bet now rests on translation. Knit's clinical deployments and shared developer foundation depend on converting an early predictive edge into routing and operational tools that hold up across institutions with different clinicians, resources and practice patterns.

Source: unite.ai

Share this article:

More from Amara Osei

Amara Osei

Show full bio

Senior reporter covering companies, deals and strategy in entrepreneurship, startups and venture funding at Business Bearings.

2,129 articles

Related articles

« Previous articleNext article »