Our pharma model, fine-tuned into yours

The Orcana model works out of the box, trained on harmonized public and licensed-partner pharma data. Your private model is the Orcana model fine-tuned on your data, owned by you and running in your cloud.

Orcana-5B and Orcana-20B in early access Orcana-30B in training Runs inside your VPC

What it's trained on

Pharma's public record, licensed partner data and our own graded work, all linked together

Public150K+

Drug labels and approvals

Public540K

Clinical trials

Public37M

Papers and abstracts

Public8.9M

Provider records

Public1.2B

Prescribing and payment rows

Public20K+

Patents and exclusivity

Partner300M

De-identified patient journeys

Partner6,000+

Health plans

Orcana310K

Graded analyses

Orcana42

Therapy-area packs

Orcana18

Validated workflows

Orcana48B

Training tokens

Partner data is licensed. Patient-level data is de-identified and never leaves the partner's governed environment in identifiable form.

The Orcana model family

Three sizes, each matched to the job the agent has to do

5Bparams

ORCANA-5B · SIGNAL

For the agents in every pod

Fast and light enough to sit inside every agent all day. Reads incoming signals, decides what matters, routes alerts and drafts the brief.

Context: 128K tokens
Runs on: 1 GPU in your VPC
Best for: Signals, triage, briefs

● Early access

20Bparams

ORCANA-20B · ANALYST

For the analysis itself

The methodologist. Turns a business question into a defensible plan: the right population, the right window, the right test. Then hands the plan to tested code to run.

Context: 256K tokens
Runs on: 2–4 GPUs in your VPC
Best for: Study design, causal, line of therapy

● Early access

30Bparams

ORCANA-30B · STRATEGIST

For the big decisions

Reasons across brands, markets and the competitive landscape for launch, portfolio, pricing and loss-of-exclusivity questions.

Context: 256K tokens
Runs on: 4–8 GPUs in your VPC
Best for: Launch, portfolio, LoE

● In training

How it scores

Smaller than frontier models, and better at pharma analytics

Internal evaluation on 1,200 held-out commercial analyses. Methodology available on request.

Orcana-20Bwith deterministic workflows

96%

Orcana-5Bwith deterministic workflows

89%

Frontier general modelwith expert prompting

62%

Frontier general modelon its own

41%

Share of analyses free of all eight error types

Your private model

Fine-tuned on your data, owned by you

Out of the box

The Orcana pharma model

Trained on: Harmonized public and licensed-partner pharma data
Knows: Pharma: labels, trials, payers, prescribing and how launches play out
Ready: Day one
Owned by: Orcana, licensed to you
Gets better with: New sources and Orcana's own research

Yours alone

Your private model

Trained on: The Orcana model, plus your sales, CRM, hub, specialty pharmacy and media data
Knows: Your brands, markets, definitions and past decisions
Ready: Once fine-tuned on your data, in your cloud
Owned by: You, including the IP
Gets better with: Every decision loop you run

Your data never trains anyone else's model
Every version is registered, validated and reversible

Get started

Pick one decision loop. We'll have it running on your data in 7–14 days

Book a demo See the platform

Runs in your cloudOne use case first, then scaleSOC 2 Type II and ISO 27001 compliant