Healthcare AI Models: Which LLMs Can Use PHI (Nov 2026)
Healthcare AI models you can use with PHI: Claude, GPT, Gemini, DeepSeek, GLM and Kimi on 7 BAA-eligible providers under one Concentrate BAA ($499/mo).
Concentrate.ai
On this page
The healthcare AI models most teams build with today are general LLMs (Claude, GPT, Gemini, and open models like DeepSeek, GLM, Kimi and Qwen) used for clinical notes, prior authorization, coding, chart summaries and patient messages. The model only matters if you can send it protected health information (PHI). On Concentrate, 70 models across 7 BAA-eligible providers can be used with PHI under one Business Associate Agreement (BAA).
Below is every major model family you can use with PHI as of November 2026, the providers that serve each under a BAA, and how to choose.
What counts as a healthcare AI model?
Healthcare AI models fall into two groups: specialized medical models and general LLMs used for healthcare work. Specialized models, like the imaging and pathology models in Microsoft Foundry's healthcare catalog, do one narrow task. General LLMs handle the text-heavy work: reading long patient records, drafting notes and letters, coding claims and answering patient questions. Independent tests like the Artificial Analysis Healthcare and Medical Index score general LLMs on medical knowledge, long patient records, tool use and hallucination.
Which healthcare AI models can you use with PHI today?
Today, these model families can be used with PHI on Concentrate, each on providers marked BAA-eligible:
Model family
LLM Gateway
Instant access to every model
Change models, providers, add fallbacks, and more by running all AI usage through one platform.
Anthropic, Azure (older Claude also on Google Vertex AI)
Sonnet 5.5: $2 / $10
GPT (OpenAI)
GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Astra, GPT-6 Luna
OpenAI, Azure
GPT-6.1 Terra: $1 / $5
Gemini (Google)
Gemini 4 Argon, Gemini 4 Carbon, Gemini 3.8 Flash
Google Vertex AI
Gemini 3.8 Flash: $0.75 / $3.75
DeepSeek
DeepSeek V4.1 Flash 1017
Fireworks AI, Azure Fireworks, Baseten (V4.1 Flash also on Azure)
$0.30 / $1.20
GLM (Z.ai)
GLM-5.3, GLM-5.3-Flash
Fireworks AI, Azure Fireworks, Baseten
GLM-5.3-Flash: $0.15 / $0.50
Kimi (Moonshot AI)
Kimi K3.1 Pro, Kimi K3
Fireworks AI, Baseten
Kimi K3.1 Pro: $2 / $10
Qwen (Alibaba)
Qwen3.8 Max
Fireworks AI
$2 / $6
Grok (xAI)
Grok 4.7
Azure, Google Vertex AI
$2 / $6
MiniMax
MiniMax M3
Fireworks AI
$0.30 / $1.20
Examples as of November 2026, at provider rates. Older routes (Claude 4.x, GPT-5.x, Gemini 2.5 and 3.x) are covered too. See Model Fortress or the HIPAA LLM gateway page for the live list of 117 covered routes.
How do you choose a healthcare AI model?
Pick by task, then by price, and test on your own de-identified data first. A common split:
Hard clinical reasoning and long records: Claude Opus 5.5, GPT-6 Astra or Gemini 4 Argon.
Everyday notes, letters and summaries: Claude Sonnet 5.5, GPT-6.1 Terra or Gemini 3.8 Flash.
High-volume extraction and coding: Claude Haiku 5.5 ($0.10 / $0.50), GPT-6 Luna ($0.10 / $0.50), GLM-5.3-Flash ($0.15 / $0.50) or DeepSeek V4.1 Flash ($0.30 / $1.20).
Because every model sits behind the same API and the same BAA, switching models is a one-line change to the model name, not a new contract.
What about OpenAI for Healthcare and Gemini for healthcare?
OpenAI for Healthcare and Google's Gemini for healthcare each cover one company's models. Anthropic, AWS and Microsoft likewise sign BAAs for their own services. That works if you only ever want one model family. If you want Claude for long records, Gemini for cheap summaries and an open model for bulk coding, each direct route means another BAA, another contract and another bill.
Why use one BAA for many healthcare AI models?
Concentrate signs one BAA with you and holds BAAs with all 7 providers behind it: Anthropic, OpenAI, Azure, Google Vertex AI, Fireworks AI, Azure Fireworks and Baseten. Aptible's LLM Gateway, the other gateway that covers inference under its own BAA, reaches 3 providers and lists no Gemini models. On a Concentrate HIPAA organization:
Zero data retention is locked on and prompt and response bodies are not stored.
Uncovered routes are blocked, including fallbacks; if no covered route is left, the request is rejected.
Models with several covered routes, like Claude Sonnet 5.5 on Anthropic and Azure, fail over without leaving the BAA.
A signed BAA with Concentrate. With a BAA, Concentrate is $499/month, with no per-request fees and no markup on tokens. Without a BAA there is no subscription fee.
A HIPAA-enabled organization, turned on by Concentrate staff after the BAA is signed.
Routing only to providers marked BAA-eligible, which HIPAA mode enforces.
Under HHS guidance, a cloud or AI service that handles ePHI for you must sign a BAA. You still own your own HIPAA obligations. Start at concentrate.ai/compliance.
FAQ: healthcare AI models, HIPAA and BAAs
What is the best AI model for healthcare?
It depends on the task. Frontier models like Claude Opus 5.5, GPT-6 Astra and Gemini 4 Argon handle hard reasoning; cheaper models like Claude Haiku 5.5 and GLM-5.3-Flash handle bulk work. All of them can be used with PHI on Concentrate under one BAA.
Are healthcare AI models HIPAA compliant?
No model is HIPAA compliant on its own. A model can be used with PHI under a signed BAA, on a provider that is BAA-eligible, with your own safeguards in place.
Can I use open models like DeepSeek with patient data?
Yes. DeepSeek V4.1 Flash, GLM-5.3, Kimi K3, Qwen3.8 Max and MiniMax M3 run on BAA-eligible providers on Concentrate.
Does Concentrate sign a BAA?
Yes. BAA use is $499/month, with no per-request fees and no markup on tokens. Non-BAA use has no subscription fee.
Does one BAA cover every healthcare AI model?
It covers every model on a BAA-eligible provider, 70 models today. Check Model Fortress for the live list.
Blog