# Live GPU prices as JSON: our public pricing API
18 November 2026 · Product · SwissGPU
Current prices for every SwissGPU server are available as JSON. One public, read-only request returns hourly rates in CHF and indicative USD, specifications, prepaid terms and fees. You need no account and no API key.
We built it for anyone who compares GPU providers with a script instead of a browser: price trackers, procurement tools, and AI agents that evaluate options for a workload.

## At a glance

- `GET https://www.swissgpu.ch/api/v1/offers` returns each GPU model with specifications, the on-demand hourly price, prepaid terms, fees and rental links.

- Prices are in CHF, with indicative USD values next to them.

- No authentication. The endpoint reads the same live catalogue as the pricing page.

- Limit: an average of 60 requests per minute per IP address. Polling once per minute is enough.

- `GET /api/v1/pricing`, the earlier compact price map, remains supported for existing integrations.

## Every offer in one request

```
curl https://www.swissgpu.ch/api/v1/offers
```

Each entry in `offers` describes one GPU model, with what you need to judge whether a server fits a workload. This is the RTX 5090 entry on 18 November 2026, shortened to one prepaid term. Prices and exchange rates change, so always read the live values:

```
{
  "gpu": "RTX5090",
  "name": "NVIDIA RTX 5090",
  "availability": "available",
  "prepaid_terms": [
    {
      "months": 1,
      "discount_percent": 5,
      "hourly_cents": 47,
      "compute_cents": 33840,
      "hourly_usd": 0.5594,
      "compute_usd": 402.7594
    }
  ],
  "hourly_price": 0.5,
  "hourly_price_usd": 0.5951,
  "specifications": {
    "vram_gb": "32",
    "cpu_cores": "16",
    "ram_gb": "32–64",
    "storage": "1–2 TB NVMe",
    "architecture": "Blackwell"
  },
  "model_url": "https://www.swissgpu.ch/gpus/rtx-5090",
  "rental_url": "https://www.swissgpu.ch/rent/rtx5090",
  "specifications_url": "https://www.nvidia.com/en-eu/geforce/graphics-cards/50-series/rtx-5090/"
}
```

The response also carries a `fees` object, so a cost estimate does not depend on reading a web page: electricity per kWh, the installation fee, shared IPv4 per day and the Windows arrangements.

## What the numbers mean

- `hourly_price` is the on-demand rate in CHF per hour, before additional fees: `0.5` means CHF 0.50 per hour for a dedicated RTX 5090 server. Electricity is metered and billed separately, and each operating system installation has a fee.

- `prepaid_terms` lists the enabled fixed-term offers. `hourly_cents` is the discounted hourly rate and `compute_cents` the upfront compute purchase, both in integer CHF cents. An empty list means no prepaid term is enabled for that model.

- `hourly_price_usd`, `hourly_usd` and `compute_usd` are indicative US dollar amounts. Billing is always in CHF. The conversion uses the European Central Bank’s daily reference rate, which the response reports in `usd_exchange_rate`.

- `availability: "available"` means the model is commercially offered. It is not a server count or a reservation. Whether a model can be rented right now is answered by the token-protected capacity endpoint of the [customer API](https://www.swissgpu.ch/news/gpu-rental-api).

- `observed_at` is the UTC time the response was generated. Fetch again when you need current rates.

## No stale or invented prices
If the current catalogue cannot be read completely, the offers endpoint answers with HTTP 503 and a `Retry-After` header. It never substitutes a cached or zero price. A script or agent that receives a price can rely on it being the price shown on the website at that moment.

## The compact price map
`GET /api/v1/pricing` returns a single hourly rate per GPU model, in CHF and indicative USD. It remains supported for existing integrations. Where a 12-month prepaid term is enabled for a model, it returns that term’s rate; otherwise it returns the on-demand rate. Use the offers endpoint when you need to tell the two apart.

## Made to be found by agents
More and more infrastructure decisions start with a question to an AI assistant. We would rather have that assistant read our actual prices than guess them. So the site publishes an [OpenAPI schema](https://www.swissgpu.ch/api/v1/openapi.json) for these endpoints, an [llms.txt](https://www.swissgpu.ch/llms.txt) index of the important pages, and a Markdown version of every public page.

## Start here

- [Offers and availability API](https://www.swissgpu.ch/docs/offers-api): every response field, USD estimates and the rental handoff.

- [Legacy Pricing API](https://www.swissgpu.ch/docs/pricing-api): the compact price map for existing integrations.

- [Pricing page](https://www.swissgpu.ch/pricing): the same rates for people, with all billing details.

- [Customer API](https://www.swissgpu.ch/news/gpu-rental-api): rent and manage servers with a token.