ThunderPeak API documentation

Quickstart

1. Sign up at console.thunder-peak.com and create an API key (sk-tp-…).
2. Point your tool at the base URL below.
3. Send your first request.

Try it with curl (OpenAI-compatible chat):

Tip: query your live model list at any time with GET /v1/models using the same key — the catalogue is always current and per-key.

Protocols & endpoints

All three wire formats share one key and one credit balance. Request bodies and responses are byte-compatible with the upstream APIs — point any official SDK at the matching base URL.

• Chat Completions — OpenAI-format messages, function calling, streaming via SSE. Use with OpenAI SDKs, Cherry Studio, most tools.
• Anthropic Messages — x-api-key header + anthropic-version. Use with Claude Code and Anthropic SDK.
• Responses — the modern OpenAI agent protocol used by Codex CLI/extension.

Video generation

HappyHorse 1.1 generates video from text, an image, or a reference clip. It uses the same API key and the same balance as your chat models.

• Web UI: open /video, paste your key, describe the shot and generate.
• API: POST /api/video/generate with {key, model, prompt, size, duration} → returns a task_id; poll POST /api/video/status until state = SUCCEEDED and read the url.
• Models: happyhorse-1.1-t2v (text), happyhorse-1.1-i2v (image), happyhorse-1.1-r2v (reference).
• Pricing: 720P $0.38 per 5-second clip · 1080P $0.68 per 5-second clip (10s = ×2). Charged when the job is accepted; failed jobs are refunded automatically.
• Notes: generation usually takes 1–3 minutes; output links are temporary, download promptly.

Tool configuration

Billing & tokens

• Credits never expire. Top up with USDT on TRC-20 / BSC / Solana (min $10): copy the address on the Plans page, send the transfer, then paste your TXID back on that page — your balance is credited automatically, usually within a minute.
• New accounts receive $2.50 of free test credit after email verification — try the gateway before buying credits.
• Per-request metering covers input, output, cache-write, cache-read and thinking tokens. Streaming costs the same as non-streaming.
• Listed prices: Claude & GPT at 66% of official rates; DeepSeek & Kimi K3 88%; GLM-5.3 80%, GLM-5.2 70%; Qwen at official rates. Discounts are a standing policy, not a promo.
• Your balance and per-key usage are visible in the console; query balance server-side with GET /v1/balance.

Vision

qwen3-vl-plus understands images: describe screenshots, read documents, OCR, or analyse photos. It uses the same API key and balance as your chat models.

• Call format (OpenAI-compatible):

  POST /v1/chat/completions
  {
    "model": "qwen3-vl-plus",
    "messages": [{
      "role": "user",
      "content": [
        {"type": "text", "text": "Describe this image"},
        {"type": "image_url", "image_url": {"url": "https://example.com/photo.jpg"}}
      ]
    }]
  }

• Pricing: $0.32 / $1.28 per 1M tokens (official rates). Images are billed as image tokens (roughly 100–1200 tokens per image depending on size).
• Image sources: use a publicly reachable URL, or inline the image as a base64 data URL (data:image/png;base64,...).
• Image proxy (recommended for images hosted abroad): our inference region sits in mainland China, so overseas image hosts may time out. Wrap any external URL with our proxy and it works everywhere:
  https://thunder-peak.com/img-proxy?u=<URL-encoded image url>
  Example: https://thunder-peak.com/img-proxy?u=https%3A%2F%2Fpicsum.photos%2F400%2F300
  The proxy fetches the image once, caches it, and serves it from our domain — sub-second on repeat calls.

Errors & rate limits

Keys default to a conservative requests-per-minute limit; raise or lower it per key in the console. During upstream outages the gateway returns 5xx while GET /v1/models and /healthz stay green — that distinction means the problem is upstream, not your key.