Epic Cosmos
Summary New Core Watchlist Market Weak signals Manual checks Gaps & log Sources
Executive summary
- Epic Cosmos now reports 304M patients and 20.7B encounters (up from 300M/16.3B in Aug 2025) — the largest single US EHR research corpus. ↳
- The flank holds hard: 'Access to Cosmos cannot be purchased' — members-only, contribution-to-use; pharma/life-sciences/payers cannot license it. ↳
- Epic shipped Comet/CoMET, a generative EHR foundation model (118M patients, 115B events) built with Yale + Microsoft — among the category's deepest AI. ↳
- Cosmos appears in ~65 publications in 90 days — heavy academic adoption as a research substrate. ↳
- For Inovalon, Cosmos sits on the opposite axis from payer-grade RWD: clinical EHR data, no payer/claims constructs, not commercially licensable. ↳
- Comet raises the category AI bar (with IQVIA Med-R1, Truveta TLM); the answer is payer-grade reasoning, not cloning an EHR model. ↳
New since last run
factnew Epic Cosmos reports 304M patients, 20.7B encounters and 10B face-to-face visits across 2,133 hospitals and 48.8K clinics, up from ~300M patients / 16.3B encounters stated in Aug 2025.
factnew Cosmos states 'Access to Cosmos and Cosmos data cannot be purchased'; direct access is limited to clinicians affiliated with and approved by a participating member organization on a contribution-to-use basis, so external pharma, life-sciences and payers cannot independently license it.
factnew Epic launched Comet, a generative 'Cosmos Medical Event Transformer' (CoMET) family of decoder-only transformer models pretrained on 118M patients / 115B medical events (151B tokens) with Yale and Microsoft, which matched or beat most single-task predictive models across 78 early evaluations.
factnew Cosmos appears in roughly 65 PubMed/NIH-RePORTER-indexed outputs in the trailing 90 days, spanning broad epidemiology (e.g., longitudinal maternal-infant cohorts, antidepressant prescribing trends, telehealth resource-use, SDOH Z-code use), evidencing heavy academic adoption as a research substrate.
factnew Advocate Health transitioned to Epic Cosmos in Feb 2026, expanding the contributing member base behind the dataset.
inferencenew Cosmos is an EHR-scale + generative-AI structural threat to RWD vendors generally, but does NOT contest Inovalon's payer-grade lane: it is clinical EHR data with no payer/claims constructs (risk adjustment, HEDIS/Stars, RADV provenance, enrollment denominators, reimbursement economics) and 'cannot be purchased', while Epic's only commercial life-sciences motion is trial matchmaking (Discovery), not a licensable RWD product.
inferencenew Comet pushes the category AI bar alongside IQVIA Med-R1 and Truveta TLM — a generative foundation model trained on the largest US EHR corpus and externally published (Yale/Microsoft/arXiv), reinforcing that Inovalon's defensible AI is reasoning over payer-grade structured data rather than cloning an EHR foundation model.
Core competitors
Watchlist headlines
Epic Cosmos
factnew Epic Cosmos reports 304M patients, 20.7B encounters and 10B face-to-face visits across 2,133 hospitals and 48.8K clinics, up from ~300M patients / 16.3B encounters stated in Aug 2025.
factnew Cosmos states 'Access to Cosmos and Cosmos data cannot be purchased'; direct access is limited to clinicians affiliated with and approved by a participating member organization on a contribution-to-use basis, so external pharma, life-sciences and payers cannot independently license it.
factnew Epic launched Comet, a generative 'Cosmos Medical Event Transformer' (CoMET) family of decoder-only transformer models pretrained on 118M patients / 115B medical events (151B tokens) with Yale and Microsoft, which matched or beat most single-task predictive models across 78 early evaluations.
factnew Cosmos appears in roughly 65 PubMed/NIH-RePORTER-indexed outputs in the trailing 90 days, spanning broad epidemiology (e.g., longitudinal maternal-infant cohorts, antidepressant prescribing trends, telehealth resource-use, SDOH Z-code use), evidencing heavy academic adoption as a research substrate.
factnew Advocate Health transitioned to Epic Cosmos in Feb 2026, expanding the contributing member base behind the dataset.
inferencenew Cosmos is an EHR-scale + generative-AI structural threat to RWD vendors generally, but does NOT contest Inovalon's payer-grade lane: it is clinical EHR data with no payer/claims constructs (risk adjustment, HEDIS/Stars, RADV provenance, enrollment denominators, reimbursement economics) and 'cannot be purchased', while Epic's only commercial life-sciences motion is trial matchmaking (Discovery), not a licensable RWD product.
inferencenew Comet pushes the category AI bar alongside IQVIA Med-R1 and Truveta TLM — a generative foundation model trained on the largest US EHR corpus and externally published (Yale/Microsoft/arXiv), reinforcing that Inovalon's defensible AI is reasoning over payer-grade structured data rather than cloning an EHR foundation model.
Market & demand signals
No demand-side findings this run.
Weak signals
Single-signal inferences — interpret with caution.
Manual checks
Web/LinkedIn search Epic careers for Cosmos / CoMET / research-data roles to gauge investment pace; Epic uses Avature behind careers.epic.com with no public JSON. — https://careers.epic.com/
First snapshot stored this run; next run's page-diff will show patient/encounter growth deltas automatically. — https://cosmos.epic.com/about/
Track whether Comet/CoMET ever leaves the member virtual-lab for commercial or pharma use — that would change the threat assessment materially. — https://arxiv.org/html/2508.12104v1
Coverage gaps & run log
Coverage gaps
| Company | Category | Reason |
|---|---|---|
| Epic Cosmos | trials | all 2 scripted attempt(s) empty/failed (cap 2 for watchlist tier) |
| Epic Cosmos | regulatory | all 2 scripted attempt(s) empty/failed (cap 2 for watchlist tier) |
| Epic Cosmos | financial | all 1 scripted attempt(s) empty/failed (cap 2 for watchlist tier); Epic is private, no EDGAR/earnings |
| Epic Cosmos | jobs | no public ATS JSON (Avature behind careers.epic.com); deferred to Manual checks per skill quirk rather than spending watchlist web-search cap |
Query log
| Source | Query | Results | Status |
|---|---|---|---|
| clinicaltrials | ct sponsor=Epic Cosmos | 0 | ok |
| clinicaltrials | ct term=Epic Cosmos | 0 | ok |
| openfda-510k | 510k applicant=Epic Cosmos | 0 | ok |
| openfda-pma | pma applicant=Epic Cosmos | 0 | ok |
| pubmed | pubmed Epic Cosmos | 52 | ok |
| nih_reporter | reporter text=Epic Cosmos | 13 | ok |
| edgar-fts | edgar fts Epic Cosmos | 0 | ok |
| pages | pages Epic Cosmos | 2 | ok |
| web | WebFetch cosmos.epic.com/about (scale + access model) | 1 | ok |
| web | "Epic Cosmos" generative AI model 2026 Comet research | 10 | ok |
| web | "Epic Cosmos" life sciences pharma external access dataset licensing | 10 | ok |
| web | "Epic Cosmos" 2026 patients encounters dataset size announcement | 10 | ok |
Sources
Epic Cosmos
- https://cosmos.epic.com/about/
- https://cosmos.epic.com/request-access/
- https://hitconsultant.net/2025/09/03/epic-launches-comet-a-new-ai-platform-to-predict-patient-health-journeys/
- https://arxiv.org/html/2508.12104v1
- https://www.beckershospitalreview.com/healthcare-information-technology/ehrs/epic-unveils-ai-model-for-clinical-foresight/
- https://pubmed.ncbi.nlm.nih.gov/42282186/
- https://ctsi.wakehealth.edu/announcements/2026/02/advocate-transitions-to-epic-cosmos
- https://www.epic.com/software/life-sciences/