Quickstart: compare models in the playground

In this quickstart, you send a prompt to a Mistral model in the Studio›Studio ↗ playground, change the temperature to see its effect, and compare the answer of a second model. It takes about 5 minutes and needs no code. Use it to choose a model and its parameters before you call the API from your application.

• A Mistral account ↗ with access to Studio.

1. Open the Studio console.
2. In the Studio navigation, select Playground.

The playground opens a chat interface where you can send prompts to the available models.

1. In the Model list, select mistral-small-latest.

2. Enter the following prompt:

   Explain the difference between supervised and unsupervised learning in three sentences.

   Explain the difference between supervised and unsupervised learning in three sentences.

3. Select Send, or press Enter.

The model's answer appears in the chat panel.

The temperature controls how much the output varies: a low value gives focused, repeatable answers, a high value gives more varied ones.

1. In the parameters panel, set Temperature to 0.1, then send the same prompt.
2. Set Temperature to 0.9, then send the same prompt again.

The parameters panel also has Max tokens, which limits the length of the answer, and Top P, an alternative to the temperature for controlling diversity.

1. In the Model list, select mistral-large-latest.
2. Send the same prompt.

• Three sentences that contrast the two kinds of learning, from each model.
• At 0.1, an answer that changes little when you send the prompt again. At 0.9, more variation in wording and examples.
• Differences in length, detail, and style between mistral-small-latest and mistral-large-latest.

Note the model and the temperature that suit your task: you pass the same values as model and temperature when you call the API.