The Playground is where you run real requests against a model from the browser. It is paid exactly like an API call — by your Agent plan when you have one with room, otherwise from your balance — so it is the fastest way to try a model, compare protocols, and copy working code.
New here? Start with What is Tokener? and Quickstart without code.
The controls
- Model — choose which model answers. The list is your live catalogue; the featured model is selected first.
- Protocol — choose how the request is formatted: OpenAI-compatible (Chat Completions), OpenAI Responses, or Anthropic-compatible (Messages). All three reach the same models with the same key. A control a protocol cannot express is disabled with a hint: under OpenAI Responses, seed and stop are disabled; under Anthropic-compatible (Messages), JSON output and seed are disabled.
- Reasoning — pick the model's reasoning mode where it has one.
- Max tokens — cap the length of the reply.
- System prompt — set standing instructions for the model (in the settings tabs).
Scenarios
The Scenarios chips fill the prompt and open the matching settings for a common task: Chat, Tool calling and JSON output. Long context is a separate toggle in the header that pads the prompt to a chosen size so you can try a large request.
Attachments
Use Add files to attach up to 4 files to a message. Accepted kinds and limits:
- Images up to 4 MB.
- PDF up to 10 MB (up to 10 pages).
- Word, Excel and PowerPoint up to 10 MB (Excel up to 200 rows per sheet; PowerPoint up to 8 slides).
- Video up to 25 MB (up to 8 key frames).
- Audio up to 8 MB (a recording up to 60 seconds; a file up to 600 seconds).
PDF pages, Office documents and video frames are converted in your own browser, and audio is transcribed in the browser when the model cannot take audio directly — so the original files are not uploaded, only the converted text or images (or, for a model that accepts them, the media parts).
Sending, stopping and stats
Press Send to run the request; the reply streams in. Stop ends a reply early. A stopped or failed request appears on your Usage page with the status Failed and a cost of US$0.00. Under each finished reply the stats show the token counts and the cost in US dollars — the same figures your Usage page records. With an Agent plan, the stats line ends with Agent plan when the plan covered the reply, or Balance when your balance paid for it.
When a model is temporarily unavailable
When our automatic checks cannot reach a model, it is marked Temporarily unavailable: the Model menu shows it greyed with "(temporarily unavailable)", and it cannot be selected. Pick another model in the meantime. The mark goes away by itself as soon as a check succeeds, usually within minutes. If you were already chatting with that model, your next request shows a message asking you to choose another model, and nothing is charged.
Why does the Playground ask before using some models?
By default your requests go only to model suppliers that commit to not storing request content and not using it for training (the Data handling setting). Sometimes a model is temporarily available only from suppliers without that commitment. The Model menu then shows it with "(limited suppliers)", and selecting it opens the question Use suppliers without a no-storage commitment?:
- Continue uses that model for the rest of this Playground session, with those suppliers allowed for its requests only. They may store your inputs and may process requests in other countries, under the laws of those countries. Reloading the page or opening a new tab asks again.
- Choose another model keeps your previous model.
A request sent after Continue is billed at the current price. To allow those suppliers for every request instead, turn on "Also allow suppliers without that commitment" in Settings → Data handling. If your account already allows them, the Playground never asks.
Copying the request and the code
- Request JSON shows the exact request body the Playground will send, for the selected protocol and settings.
- Get Code generates a ready-to-run snippet for the selected protocol (OpenAI-SDK or Anthropic-SDK code), using your current model and settings. Its tabs also include a Vercel AI SDK sample on the OpenAI-compatible (Chat Completions) protocol and a Claude Code setup on the Anthropic-compatible (Messages) protocol.
- Clear chat empties the current conversation.
Getting help
The support chat (the bubble on the site, and the Support page) answers questions about your account, billing, the API, models, the Playground, pricing and policies, and can read these guides. It is read-only: it cannot change your account, issue refunds, or reveal a key. For anything else, email [email protected] for general and privacy questions, or [email protected] for account access problems.