Dashboard → Playground lets you try models in the browser. It uses your API key and calls the live API, so a successful request is billed the same way as an SDK or curl call, and shows up in usage and logs.
Playground is not a free sandbox. Every successful call is charged against your prepaid balance.
Before you start
- A registered, email-verified OctaRouter account with available balance. If you are low on credits, top up first.
- At least one API key (Dashboard → API keys).
- Pick a model in Playground, or click Try in Playground on the Models page.
Model types
Choose the type that matches what you want to do. Billing can differ by model — see Models for detailed pricing, then check the actual charge after you run.
| Type | Endpoint | How you are billed |
|---|---|---|
| Chat | Responses, Chat Completions, Anthropic Messages, or Gemini Generate Content | Tokens (input / output; some models also bill cache or reasoning tokens) |
| Image | POST /v1/images/generations or POST /v1/images/edits | Image tokens, or per image (1 image per request) |
| Transcribe | POST https://api.octarouter.com/v1/audio/transcriptions | Audio duration (seconds), or audio tokens |
| Speech | POST https://api.octarouter.com/v1/audio/speech | Characters actually synthesized — not the length of the text box |
Vision chat models still use Chat and token billing. They are not image-generation models.
Route groups
A model can be published in more than one route group, such as Default and Flex. Playground shows a group switch next to the model name:
| Group | Model ID sent on the request |
|---|---|
| Default | model-id |
| Any other group | model-id:group, for example deepseek-v3.2:flex |
A model that is only published outside Default gets that suffix automatically. After you pick a group, the protocol list is limited to protocols that group supports. See Flex route group.
After you run a request
See Models for detailed prices, group discounts and billing rules. The Playground workspace focuses on testing and does not repeat pricing details. The model picker keeps standard reference rates to help you choose.
Routing, usage tiers and cache hits may affect the final charge. After the run, Playground shows the actual charge with six decimal places; nonzero amounts below $0.000001 appear as <$0.000001. The same record also appears under Dashboard → Logs.
If you cancel, or the call does not complete, it is usually not charged.
If you see an insufficient-balance error, top up and try again.
To call the same endpoints from your app, see API protocols.
Protocols and endpoints
In Parameters, choose a protocol and a request endpoint. Options come from the model catalog and are filtered by workspace type and the protocols available in the current route group. Changing the model, group, or protocol resets the selection to a valid endpoint. Sending is disabled when the catalog explicitly publishes no usable endpoints. Older gateways without endpoint metadata retain the default endpoint for each protocol.
Chat supports OpenAI Responses (/v1/responses) and Chat Completions (/v1/chat/completions), Anthropic Messages (/v1/messages), and Gemini Generate Content (/v1beta/models/{model}:generateContent). Only published endpoints are offered. The actual request path appears below the endpoint selector in Parameters. Responses supports conversation history, image input, and streaming text.
Image models expose generation and editing endpoints when published. For image edits, upload a source image and describe the requested changes. Transcription and speech use /v1/audio/transcriptions and /v1/audio/speech respectively.