LLM & Media API

One API key, one balance, one usage log — 370+ models covering text, images, video, music, speech and restoration.

There are two ways in:

1. Native GPUniq API (/v1/llm/*) — wrapped responses with USD cost on every call, server-side chat sessions, the generations library, and the async job surfaces for every media kind.
2. Drop-in compatibility — /v1/openai/* replaces api.openai.com/v1, and /v1/messages is a native Anthropic Messages endpoint. Claude Code, Cursor, Continue.dev, Aider, LiteLLM and the official SDKs work without code changes.

Where to go

Your first call

curl https://api.gpuniq.com/v1/llm/chat/completions \
  -H "X-API-Key: gpuniq_your_key" \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [{"role": "user", "content": "Explain how transformers work"}],
    "model": "claude-haiku-4-5"
  }'

Omit model and you get claude-haiku-4-5 — fast, cheap, strong at code.

Authentication & limits

API keys are created in the web dashboard under LLM API Keys. A key starts with gpuniq_ and works on every surface: marketplace, instances, volumes and every model endpoint here.

A dashboard JWT works in place of a key on native routes.

Rate limit: 120 requests/minute per key, sliding window. It is a default, not a ceiling — accounts can be raised, and a raised account limit cascades to its keys. The Python SDK retries automatically when it hits the limit.

# 403, error code 1010 — blocked at the edge, never reaches the API
urllib.request.urlopen(urllib.request.Request(url, data=body))

# Works — requests sends its own User-Agent
requests.post(url, json=payload, headers={"X-API-Key": KEY})

# Works — urllib with an explicit User-Agent
req = urllib.request.Request(url, data=body, headers={
    "X-API-Key": KEY,
    "Content-Type": "application/json",
    "User-Agent": "my-app/1.0",     # <- the one line that matters
})

Long generations & streaming

The edge proxy closes inbound connections after ~100 seconds of streaming silence. A non-streaming request asking for max_tokens > 4096 is not refused: GPUniq streams from the upstream and assembles the reply itself, flushing whitespace keep-alive bytes to you until the final JSON is ready (the response carries X-GPUniq-Long-Generation: keepalive). On that path an upstream failure arrives as HTTP 200 whose body is {"error": …} — check for it. For long replies, prefer "stream": true or the job-based long-poll API.

Reasoning models (Gemini 3 Pro, DeepSeek R1, o3, Claude Opus thinking) burn tokens on hidden chain-of-thought before the visible reply, so they need extra max_tokens headroom — see Long generations for the full streaming / job-based / reasoning-token recipe.

Errors

Every failure returns a stable OpenAI error envelope with a structured code you can branch on — insufficient_balance, model_not_found, rate_limit_per_key, provider_unavailable, etc. See the Error reference for the complete catalog, recovery strategies, and the native vs. OpenAI-compat envelope shapes.

{
  "error": {
    "message": "Unknown model 'gpt-9'. Fetch /v1/openai/models for the current catalog.",
    "type": "invalid_request_error",
    "code": "model_not_found"
  },
  "status_code": 400,
  "request_id": "…"
}

Where delivered media lives

Generated images and videos are mirrored to GPUniq storage: the response carries both the upstream URL and a GPUniq-hosted copy, kept for 7 days. That copy is a delivery buffer, not an archive — download the bytes if you need them to last. Music and audio are not mirrored; download those as soon as the job completes. Details: Where delivered media lives.