GPU Pricing Data API
Read-only access to regularly-tracked cloud GPU pricing data. 5 endpoints, standard JSON envelope, API key auth. Free tier: 100 requests/hour.
No-Key Endpoint (Attribution Required)
GET /api/v1/prices/public — no API key required. 60 req/hr per IP. Free with attribution.
curl "https://gpuadvisor.com/api/v1/prices/public?gpu=H100" # Optional filters: # ?gpu=H100 — GPU model substring # ?provider=Lambda — provider name
{
"meta": {
"source": "gpuadvisor.com",
"license": "Free with attribution — link to gpuadvisor.com required",
"last_tracked": "July 1, 2026",
"version": "v1",
"rate_limit": "60 requests/hour per IP. Use X-API-Key for higher limits."
},
"providers": {
"AWS": [
{
"gpu": "H100 80GB",
"instance": "p5.48xlarge",
"on_demand_per_gpu_hr": 6.88,
"spot_per_gpu_hr": 2.475,
"reserved_1y_per_gpu_hr": 4.8125,
"vram": "640 GB (8×80)",
"region": "us-east-1"
}
],
"Lambda": [ ... ],
"CoreWeave": [ ... ]
}
}Attribution requirement: display Data via GPUAdvisor — gpuadvisor.com wherever this data is shown to end users. For higher limits or the full snapshot history API, use a keyed endpoint below.
Keyed API — Quick Start
curl -H "X-API-Key: YOUR_KEY" \ "https://gpuadvisor.com/api/v1/prices?gpu=H100+80GB"
Authentication
Pass your API key in the X-API-Key request header. Every request must include a valid key.
All responses include rate-limit headers: X-RateLimit-Limit, X-RateLimit-Remaining, and X-RateLimit-Reset (Unix epoch).
Endpoints
GET /v1/index (median series) was retired on August 18, 2026. Callers receive 410 Gone. The GPU Price Index is a 25th-percentile CPI-style composite — read it at /gpu-price-index or via the MCP get_price_index tool.
/v1/gpusList all tracked GPU models with provider counts and first/last tracked dates.
curl -H "X-API-Key: YOUR_KEY" \ "https://gpuadvisor.com/api/v1/gpus"
/v1/providersList all tracked cloud providers with GPU models and regions covered.
curl -H "X-API-Key: YOUR_KEY" \ "https://gpuadvisor.com/api/v1/providers"
/v1/pricesCurrent normalized per-GPU $/hr from the latest snapshot. Optionally filter by GPU model, provider, or a specific snapshot date.
gpu—Filter by GPU modele.g. H100+80GBprovider—Filter by providere.g. CoreWeavedate—Snapshot date (default: latest)e.g. 2026-06-20curl -H "X-API-Key: YOUR_KEY" \ "https://gpuadvisor.com/api/v1/prices?gpu=H100+80GB"
/v1/historyTime-series price history for a GPU model, grouped by provider. Requires 'gpu' parameter.
gpurequired—GPU model — REQUIREDe.g. H100+80GBprovider—Filter to one providere.g. AWSfrom—Start date inclusive (YYYY-MM-DD)e.g. 2026-01-01to—End date inclusive (YYYY-MM-DD)e.g. 2026-12-31curl -H "X-API-Key: YOUR_KEY" \ "https://gpuadvisor.com/api/v1/history?gpu=H100+80GB&from=2026-01-01"
/v1/metaAPI and methodology metadata: last tracked date, snapshot count, provider list, tier limits.
curl -H "X-API-Key: YOUR_KEY" \ "https://gpuadvisor.com/api/v1/meta"
MCP Connector (Claude and MCP Clients)
The same regularly-tracked pricing is also exposed as an MCP server for Claude and other MCP-compatible clients. Three read-only tools, authless public tier.
Endpoint
https://gpuadvisor.com/api/mcpTools
get_pricingcurrent per-GPU $/hr across all tracked providers; on-demand, spot, and reserved rows kept separatecompare_providersside-by-side comparison for a specific GPU across every tracked provider on the same rubricbudget_plancompute cost estimate given GPU count, hours per month, and planning horizonAuthless public tier with attribution required. Same data, same freshness, and same terms as GET /api/v1/prices/public above. Every response includes attribution and a cadence note.
Response Format
Every response is wrapped in a standard envelope with a data field (endpoint payload) and a meta field (freshness, attribution, methodology link).
{
"data": [ /* endpoint-specific payload */ ],
"meta": {
"last_tracked": "2026-06-20",
"snapshot_date": "2026-06-20",
"pricing_basis": "On-demand per-GPU $/hr. Node prices divided by GPU count from instance VRAM spec.",
"provider_count": 9,
"methodology_url": "https://gpuadvisor.com/gpu-price-index",
"attribution": "Data via GPUAdvisor — gpuadvisor.com",
"cadence_note": "Regularly tracked — not real-time. Confirm before procurement."
}
}{
"snapshot_date": "2026-06-20",
"gpu_model": "H100 80GB",
"provider": "CoreWeave",
"region": "us-east-1",
"price_per_gpu_hr": 2.39,
"spot_per_gpu_hr": 1.85,
"reserved1y_per_gpu_hr": 2.10,
"reserved3y_per_gpu_hr": 1.75,
"pricing_basis_note": "on-demand · per-GPU rate"
}{ "error": { "code": "NOT_FOUND", "message": "...", "hint": "..." } }Rate Limits
Rate limit windows reset on a rolling 1-hour basis. The X-RateLimit-Reset header contains the reset epoch timestamp.
Attribution (Free Tier)
Free-tier integrations must display the following attribution string wherever pricing data is shown to end users:
The meta.attribution field in every response contains this string. For internal-only tooling (dashboards, data pipelines not visible to end users), attribution is encouraged but not required.
Pricing Basis & Data Freshness
All price_per_gpu_hr values are on-demand, normalized to per-GPU. Multi-GPU node prices are divided by GPU count derived from the instance VRAM spec. CoreWeave prices are quoted natively per GPU.
Data is regularly tracked — not real-time. The meta.last_tracked field shows the most recent tracking date. Confirm prices directly with providers before procurement decisions.
The GPU Price Index at /gpu-price-index is the single canonical index — a 25th-percentile CPI-style composite, not a median.
OpenAPI Spec
A static OpenAPI 3.1 spec is available for import into Postman, Insomnia, or any OpenAPI-compatible toolchain.
/openapi.jsonGet an API Key
Email us with a brief description of your project. Free keys are issued within 1 business day.
[email protected]