Skip to main content
GET
List GPU types

Authorizations

Authorization
string
header
required

Runpod API key authentication. Generate an API key in the Runpod console and send it in the Authorization header as Bearer <api_key>. Keys are scoped to the permissions granted when created; requests may return 403 when a valid key lacks access to the requested resource or action.

Query Parameters

include
enum<string>[]

Comma-separated optional expansions. Supported value today: AVAILABILITY. This may expand with more include values in the future.

Maximum array length: 1

Catalog include expansion. Only AVAILABILITY is supported today; additional include values may be added in the future.

Available options:
AVAILABILITY
product
enum<string>[]

Comma-separated availability product contexts. Supported values: POD, CLUSTER, SERVERLESS. Required with include=AVAILABILITY, and valid only with it (400 either way). There is no default: the same GPU type can be scarce for pods and plentiful for serverless, so the context has to be stated rather than assumed.

Catalog product availability context. Availability is product-specific, so this is required whenever availability is requested.

Available options:
POD,
CLUSTER,
SERVERLESS
count
integer
default:1

GPU count for availability and lowest-price calculations. Valid only with include=AVAILABILITY. Defaults to 1.

Required range: x >= 1
cloud
enum<string>

Cloud type for availability and lowest-price calculations. Valid only with include=AVAILABILITY. Supported values: SECURE, COMMUNITY. Upstream default when omitted: SECURE. GPU availability cloud filter.

Available options:
SECURE,
COMMUNITY
countryCodes
string[]

Comma-separated ISO 3166-1 alpha-2 country codes, uppercase, to constrain availability to — e.g. FR or FR,DE. Values within this filter use OR semantics. Valid only with include=AVAILABILITY (400 otherwise); a malformed entry is a 422. Scopes availability, lowest-price calculations and the dataCenters array to those countries, so a listed data center outside them is omitted rather than returned with availability NONE. On the list endpoint a GPU type with no data center in those countries drops out entirely; the single-GPU endpoint still returns the requested type, with availability NONE and dataCenters omitted, so a 404 keeps meaning the GPU type does not exist. Read the NONE on availability rather than the absence of dataCenters, which is also absent when availability was not requested.

Pattern: ^[A-Z]{2}$
cudaVersions
string[]

Comma-separated CUDA versions to scope availability and lowest-price calculations to, matched exactly. Format: major.minor, e.g. 12.8 — a bare major is rejected here because it identifies no version. Valid only with include=AVAILABILITY (400 otherwise) and mutually exclusive with minCudaVersion (400 if both are sent); a malformed entry is a 422. Also narrows the returned cudaVersions array; omit it to enumerate every version offered.

Pattern: ^\d+\.\d+$
minCudaVersion
string

Lowest acceptable CUDA version to scope availability and lowest-price calculations to, compared numerically. Format: integer major or major.minor, e.g. 12 or 12.1 — unlike the gpu.minCudaVersion body field on pod and endpoint create, a bare major is accepted here and means any release of that major, because this filter only widens a read. Valid only with include=AVAILABILITY (400 otherwise) and mutually exclusive with cudaVersions (400 if both are sent); a malformed value is a 422. Use this for an open-ended floor and cudaVersions for an exact set.

Pattern: ^\d+(\.\d+)?$

Response

OK

gpus
object[]
required
Last modified on August 19, 2026