# GPU Market

> What GPU compute costs right now. Live floors and supply across 36 providers. Each listing is scored 34% availability, 33% reliability, 33% price. Free. No key needed.

Data as of: 2026-11-11T11:17:21.048Z
Providers: Vast.ai, RunPod, Akash, Clore.ai, Cudo Compute, io.net, Koyeb, Thunder Compute, Akamai Linode, Microsoft Azure, Oracle Cloud, AWS, OVHcloud, Lium, Exoscale, Nosana, GPU.ai, SimplePod, IMWT (via Shadeform), Massed Compute (via Shadeform), Lambda (via Shadeform), Crusoe (via Shadeform), Paperspace (via Shadeform), Hyperstack (via Shadeform), Boost Run (via Shadeform), Latitude.sh (via Shadeform), Scaleway (via Shadeform), Voltage Park (via Shadeform), Denvr (via Shadeform), Excess Supply (via Shadeform), Horizon (via Shadeform), Verda (via Shadeform), DigitalOcean (via Shadeform), Nebius (via Shadeform), Vultr, Phyntec (via Shadeform)

## Quick start

GET https://gpumarket.cc/api/v1/search?gpu=h100&count=1&min_reliability=0.95
GET https://gpumarket.cc/api/v1/compare?gpu=h100
GET https://gpumarket.cc/api/v1/snapshot?gpu=h100&count=8
GET https://gpumarket.cc/api/v1/history?gpu=h100
GET https://gpumarket.cc/api/v1/availability?gpu=h100
GET https://gpumarket.cc/api/v1/status
GET https://gpumarket.cc/api/v1/listings/{listing_id}/booking

## MCP

Endpoint: https://gpumarket.cc/mcp (Streamable HTTP, no auth)
Tools: search_gpu_nodes, compare_gpus, rank_providers, get_market_snapshot, get_price_history, get_availability, get_node_sizes, get_availability_index, get_data_status, list_gpus, get_gpu, list_providers, list_directory, prepare_booking, list_cluster_offers, request_capacity, report_booking_outcome

## Search parameters

- gpu: GPU model, such as h100, rtx-4090 or a100.
- count: GPUs needed. Above one node, cluster and reserved offers are returned in contract_offers.
- region: americas | europe | asia-pacific
- provider: vast | runpod | akash | clore | cudo | ionet | koyeb | thunder | linode | azure | oracle | aws | ovh | lium | exoscale | nosana | gpuai | simplepod | imwt | massedcompute | lambda | crusoe | paperspace | hyperstack | boostrun | latitude | scaleway | voltagepark | denvr | excesssupply | horizon | verda | digitalocean | nebius | vultr | phyntec
- min_price: Minimum USD per GPU-hour.
- max_price: Maximum USD per GPU-hour.
- min_reliability: Excludes listings with no reliability data.
- min_vram: Minimum GPU memory in GB.
- in_stock: true = confirmed in stock only. false = everything. Omit to hide sold-out listings and keep listings with unknown stock.
- include_interruptible: Include listings the provider can reclaim during a rental.
- term: Commitment. With a term, only listings that publish a price for it are returned, priced at that term. Cluster and reserved list prices come back in contract_offers.
- sort: Defaults to fit when count is set, otherwise best. fit = closest to the requested GPU count.
- limit: integer

## Rules

- Listings change often. Check data_as_of and call booking before you rent.
- null means the provider does not publish that value.
- The provider bills you directly. Never send provider keys to GPU Market.
- Page requests, queries and booking events are logged to measure GPU demand. No IP addresses or identities are logged.

## Pages

- [API spec](https://gpumarket.cc/openapi.json)
- [Clusters and reserved terms](https://gpumarket.cc/clusters.md), API: https://gpumarket.cc/api/v1/clusters
- Request capacity nobody lists: POST https://gpumarket.cc/api/v1/capacity-requests (or the request_capacity tool)
- [Compare GPUs](https://gpumarket.cc/compare.md)
- [Cheapest GPUs](https://gpumarket.cc/cheapest-gpu.md)
- [Spot prices](https://gpumarket.cc/spot-gpu-prices.md), [CSV](https://gpumarket.cc/spot-gpu-prices/data.csv)
- [Picks by use case](https://gpumarket.cc/picks.md)
- [All listings CSV](https://gpumarket.cc/data/listings.csv)
- GPU pages: https://gpumarket.cc/gpu/{model}, /gpu/{model}/{4x|8x}, /gpu/{model}/{region}, /gpu/{a}-vs-{b}
- Provider comparison: https://gpumarket.cc/compare/{a}-vs-{b}
- [All GPUs](https://gpumarket.cc/gpu.md)
- [Providers ranked by availability, reliability and price](https://gpumarket.cc/providers.md), API: https://gpumarket.cc/api/v1/providers/ranking?gpu=h100&count=8 (or the rank_providers tool)
- Directory of every known GPU provider, including those with no live price yet: https://gpumarket.cc/api/v1/providers/directory?gpu=h100&region=europe (or the list_directory tool)
- [Method and privacy](https://gpumarket.cc/about.md)