Choose the right model for Hermes

Connect a provider and test its model against a repeatable business task.

Before you start

• A working Hermes installation.
• Access to the provider you want to use and a short test task with a known answer.

Want to do it yourself?

Difficulty

4 out of 10

Estimated completion time

20–35 minutes

Skills you’ll need: Use a model picker, authenticate a provider, and compare output quality and usage.

I’d rather get helpGet a consultationAt capacity

We’re at capacity and aren’t taking new consultation bookings right now.

How this helps your brand

Choose a model using a real work sample so recurring tasks have a predictable quality bar. Keep the sample for future comparisons.

1. 1

   Step 1: Define a test you can judge

   Prepare five sample customer questions with the correct answers. Remove personal details and account information. Use the same sample when comparing models.

2. 2

   Step 2: Choose a provider and model

   Run hermes model to open the interactive provider and model picker. Complete the provider’s authentication flow and select a model offered to your account.

   The main model handles your conversation; auxiliary model settings cover separate work such as image analysis and compression. Start by changing only the main model.

3. 3

   Step 3: Run a fresh comparison

   Start a new conversation and submit your sample. Record omissions, unsupported claims, response time, and the usage shown by your provider. Check the provider’s current billing terms rather than assuming an existing subscription covers every request.

4. 4

   Step 4: Save a workable default

   Keep the model that meets your quality needs. If authentication fails, confirm the selected provider and account before changing unrelated configuration. Rerun the same sample after switching models.

Answer these five practice questions using only the supplied reference: [reference and questions]. Mark anything the reference does not establish. Keep each answer under 60 words so I can compare results across models.