Published Updated
Agents Update: Grok 4.6, Gemini 3.7 Flash, and Our First OpenAI Model
1 min read
HealthTasks Agents now support Grok 4.6, Gemini 3.7 Flash, and GPT-5.6 Luna—the first OpenAI model in Agents. Same workflow, program choice.
HealthTasks Agents now support three current models: Grok 4.6, Gemini 3.7 Flash, and GPT-5.6 Luna.
Agents let faculty and program leaders ask questions of their own data—competency trends, evaluation patterns, voice simulation performance, placement workload—and get answers they can use in meetings, accreditation work, and faculty development. Model choice stays with your program. You are not locked into one vendor.
Our first OpenAI model
GPT-5.6 Luna is the first OpenAI model in HealthTasks Agents.
Until now, Agents ran on Grok and Gemini. Luna uses the same workflow: the same program data, the same agent experience, and the same controls your team already has. There is no separate OpenAI product to learn or another place to export data.
That matters for programs that standardize on OpenAI elsewhere and want the same option inside clinical education software—not only in general-purpose chat tools.
Models currently supported
| Model | Best for |
|---|---|
| Grok 4.6 | Questions where faculty need context and caution before acting on a ranking or trend |
| Gemini 3.7 Flash | Fast, scannable summaries for deans, coordinators, and accreditation conversations |
| GPT-5.6 Luna | Programs that want OpenAI in Agents for routine analytical questions |
We have also updated the Grok and Gemini paths to Grok 4.6 and Gemini 3.7 Flash. The product experience is unchanged. The models underneath are current.
How programs choose
Most programs pick based on who will read the answer and how much judgment the question needs.
- Gemini when the output should be easy to forward to leadership or drop into a committee packet.
- Grok when the answer might change how faculty coach students or where a program focuses improvement.
- Luna when OpenAI belongs in your stack and the question is straightforward.
In early testing on the same Agents question—strongest and weakest CJMM criteria in voice simulations—all three reached the same headline ranking. They differed in how they wrote the answer and what caveats they included. That is why model choice remains a workflow decision, not a single default for every program.
For more on how we think about model choice, see our earlier benchmark overview.
Get started
Grok 4.6, Gemini 3.7 Flash, and GPT-5.6 Luna are available in HealthTasks Agents. If you are evaluating how agentic clinical education software should work with real program data, we would like to show you the product.