Docs / Guides
Playground and compare
Try a model in the dashboard without an API key, compare up to 3 models on one prompt, and see what each run costs.
Last updated
What the Playground is
The Playground is a chat page in the dashboard. You can try a model before you write code. It uses your signed-in session, so you do not need an API key.
Chat models are available now. The Image, Video, Audio and Embeddings tabs are marked Soon and open when they are ready.
Chat with one model
| Setting | What it does |
|---|---|
| Model | A chat model from the catalog |
| System prompt | Optional instructions sent as the system message |
max_tokens | The most output tokens the answer may use |
| Advanced | temperature, top_p, stop, seed, presence_penalty, frequency_penalty. Empty fields use the model default |
| Extra parameters (JSON) | A JSON object merged into the request, for example {"logprobs": true} |
Parameters go to the model unchanged. A parameter the model does not support comes back as the model's error.
Answers stream as they are written. Each answer shows its time to first token, its tokens (input, cache, output) and its cost. The next message continues the same conversation; Clear chat starts over.
Stop ends an answer early. Tokens already sent are billed.
Compare up to 3 models
Turn on Compare up to 3 models, pick two or three models and send one prompt. Each model answers the same prompt side by side.
- Each column is its own request, with its own request ID and its own charge.
- Each column shows its cost, and Run total is the sum of the columns.
- If one column fails, the others keep going.
- Compare sends a single prompt with the system prompt and settings; it does not keep a conversation.
Cost and balance
Playground runs are charged exactly like API requests: the same balance, the same four rates and the same minimum per request. The cost shown for an answer is the usage.cost of that request. Runs are charged to your Default project and appear on the Requests page with their tokens and cost.
With no balance available for the model, the run does not start and the page asks you to top up. See Balance.
Rate limit
The Playground has a limit of 600 RPM per account, separate from the limits of your keys, and it draws on the same balance. A compare with 3 models counts as 3 requests. Over the limit, the page asks you to try again shortly. See Rate limits.
Copy as code
Copy as code shows the current request as code for the API, with the model, messages and parameters you set. The code reads your key from the CENT_AI_KEY variable and never contains your key. Create a key in Keys to run it.
What is saved
Nothing from the chat is saved. Leaving the page clears it. Requests appear on the Requests page with their metadata only. See Request data privacy.