Your account on this inference endpoint

Paste the API key you were given. It stays in this browser tab only and is sent to this server with each request, exactly as your API calls do.

API key Show account

An OpenAI-compatible chat completions API served from this host. Point any OpenAI SDK at the base URL below with your key; nothing else changes.

Base URL

…/v1

Endpoints

• POST /v1/chat/completions one JSON reply, or server-sent events with "stream": true
• GET /v1/models what is served, with the price card when one is published
• GET /v1/usage your balance and settled usage, which this page shows

Getting a key

Keys are issued by Jerry at Hyperbolic, one per person, shown once. Ask him for one and for credit on it; then paste it above to see prices, examples and a chat that uses it.

Status

The pill above is a live GET /readyz: operational when the API and its model runtime answer, not ready when the runtime is down, unreachable when this host does not answer.

Quick start

Base URL …/v1. Set TAOTIE_API_KEY to your key first; the examples never embed it.

Good to know

• Replies carry reasoning_content, the model's thinking, before content. Both count as completion tokens. A small max_tokens can be spent entirely on thinking and return an empty answer with finish_reason: "length"; give the model room (512 or more) or omit the field to use this endpoint's default output cap.
• n must be 1 or omitted: one completion per request. Bound the output with either max_tokens or max_completion_tokens; a bound above the endpoint's cap is refused with 400 request_limit_exceeded.
• A prompt prefix the engine already holds is reported in usage.prompt_tokens_details.cached_tokens and billed at the cached-input price; the rest of the prompt is billed at the input price. Repeating a long system prompt or conversation verbatim is what makes a prefix cacheable.
• With "stream": true the final chunk before [DONE] carries usage; this server always includes it, as if stream_options.include_usage were set. A stream that fails mid-way ends with a data: {"error": …} frame and no [DONE].
• Errors: 401 invalid_api_key or missing_api_key; 403 tenant_suspended; 402 insufficient_balance (type insufficient_quota: your wallet cannot cover the request's reservation at the current price, ask for credit); 404 model_not_found; 400 request_limit_exceeded; 503 model_not_priced (no price is published for the model); 503 overloaded (retry shortly). Every response carries an x-request-id header; quote it when reporting a problem.
• Charges settle asynchronously from the engine's reported token counts; a request shows under pending settlements below until then. Funds are reserved while a request is in flight and released at settlement.

By model

Model	Requests	Fresh input	Cached input	Output	Charge (USD)