One catalog.
Clear, traceable prices.

Start with a single request

This site's read-only API is available without a key. Use the source and observation time when displaying a price.

https://xpu.live/wp-json/xpu/v1/

curl 'https://xpu.live/wp-json/xpu/v1/prices?kind=gpu&provider=runpod&limit=5'

Prices include published tariffs and observed marketplace offers. They do not guarantee current GPU availability or your final bill. Unknown prices are omitted; an explicit "0" means a published zero rate.

Explore the endpoints

List endpoints accept limit (1–200, default 50) and offset (default 0). Responses include data, total, next_offset and generated_at. resource accepts a provider's exact model ID or the catalog ID.

Try the API

Select an example and send a request.

A price is more than a number

Each record describes one provider, model or GPU, and tariff. Monetary values are decimal strings. Never compare different billing units or silently add optional components.

{
  "id": "runpod:h100-sxm:secure-cloud-pods-57c27da7",
  "kind": "gpu",
  "provider": "runpod",
  "resource": "h100-sxm",
  "currency": "USD",
  "variant": "Secure Cloud Pods",
  "price_type": "list",
  "components": [{"name": "compute", "amount": "3.49", "unit": "gpu_hour"}],
  "conditions": {"rental": "on-demand", "gpu_count": 1},
  "source": {"url": "https://www.runpod.io/pricing", "method": "page"},
  "observed_at": "2026-09-28T00:00:00Z",
  "stale": false
}

• components: input, output, cache or usage charges, each with its own unit.
• conditions: tier, modality, rental term and provider-specific restrictions.
• price_type: list, starting or marketplace. Starting rates are not exact quotes; marketplace offers are observations at collection time.
• catalog_id: identifies an exact site catalog match; null or absent for unmapped resources. These rates remain available through /prices.
• source.method: api, page or page-data (structured data published with a pricing page).
• source_status: latest collector result. stale is based on observation age: 48 hours by default, or the record’s stale_after_seconds override (Vast: 3,600 seconds).

Provider coverage

This registry covers the configured collection sources, including sources without dedicated WordPress provider pages. Counts and statuses below come from the latest imported snapshot. An active collector does not imply every product is covered. Salad and Vast prices can be queried by provider ID even without a dedicated provider page. “Mapped” counts exact matches to the site catalog.

GPU sources and pricing scope

A registered provider is not necessarily returning prices. requires_key means credentials are missing; error means the latest attempt failed; review_required needs source review. Saved prices may still be returned, so inspect their timestamps and stale flag.

Freshness, behavior and limits

• Collector refresh defaults to every 6 hours while the Node service is running. Configure 5 minutes to 7 days in CLI Settings; the running service checks settings within 30 seconds. WordPress imports hourly when WP-Cron runs, which depends on site traffic.
• Last WordPress sync: 2026-10-03T16:52:22+00:00.
• A failed source retains its previous rates and reports an error. Rates become stale after their freshness window (default 48 hours; Vast 1 hour). A successful collector status does not by itself guarantee a fresh rate. Check both fields.
• Price changes are recorded in history. A successful accepted collection replaces that provider’s saved public rates; suspicious drops can require review. A missing marketplace offer may simply be outside the bounded search result.
• observed_at is the rate observation time; generated_at is the snapshot generation time. Neither is the WordPress import time. API timestamps use UTC.
• HTTP 400: invalid pagination. HTTP 404: route not found. HTTP 503: no snapshot yet.
• The local Node API uses /v1/ routes on port 4318 and limits requests to 120/minute per client. The WordPress mirror currently has no application-level quota.
• The Node API binds to loopback by default. Public deployment requires HTTPS, an API key and gateway limits. This local site is not an internet-hosted API.
• Account-specific pricing is kept private. Public pages expose public list rates, starting rates and marketplace observations. Private organization rates are excluded.

For future MCP integration, use the same versioned price records and comparison logic. The current release exposes REST; an MCP server is not enabled.

Manage collection locally

From the pricing-service directory, run:

npm run menu

• GPU providers / AI model providers: choose a provider to see its saved price table immediately; update prices, search, filter or inspect price history from there.
• Status & service management: inspect collection results and start or stop the background service.
• Settings: view required API connections, adjust the automatic collection interval and choose automatic, always-on or disabled colors.

Use arrow keys and Enter, or the displayed option numbers. CLI times use your computer’s local timezone. Store provider keys in the local pricing-service/.env file; never send them to this public API.