DeepSeek-V4-Flash

DeepSeek-V4-Flash-t4

Azure Tier 4: Global Access MIXED Retains data

Data-use posture

Training

Not used to train the provider's models

Additionally, prompts and completions are not used to train, retrain, or improve the base models.

as of 2026-06-24 View on learn.microsoft.com ->

Logging

Provider may log prompts for abuse monitoring

The abuse monitoring data store where prompts and completions are stored for human review is logically separated by customer resource (each request includes the resource ID of the customer's Foundry resource).

as of 2026-06-24 View on learn.microsoft.com ->

Retention

Provider may store flagged prompts and completions for human abuse review, and does not publish a maximum retention period; does not apply if your deployment is approved for modified or zero abuse monitoring

The value of 'false' for the 'ContentLogging' attribute appears only if data storage for abuse monitoring is turned off. Otherwise, this property will not appear in either Azure portal or Azure CLI's output.

as of 2026-10-21 Source document

Target 1: Provider API (observed) as of 2026-10-21T04:52:36.092862+00:00 Source document

The value of 'false' for the 'ContentLogging' attribute appears only if data storage for abuse monitoring is turned off. Otherwise, this property will not appear in either Azure portal or Azure CLI's output.

Contextual documentation (not current target evidence) as of 2026-08-01 Source document

Such prompts and completions can be accessed for human review only by authorized Microsoft employees via Secure Access Workstations (SAWs) with Just-In-Time (JIT) request approval granted by team managers.

Prompt caching

Not yet verified for this route

Stateful features

Not yet verified for this route

Operational metadata

Not yet verified for this route

Legal basis

Not yet verified for this route

Where requests are processed

Run by

Microsoft Azure

Models sold by Azure are also hosted by Azure and operated by Azure as part of the Foundry Models service.

as of 2026-10-08 Source: learn.microsoft.com

Processed in

Any region where the provider runs this model. Listed today: North America: Canada, United States. South America: Brazil. Europe: France, Germany, Italy, Netherlands, Norway, Poland, Spain, Sweden, Switzerland, United Kingdom. Asia: India, Japan, Singapore, South Korea, United Arab Emirates. Oceania: Australia. Africa: South Africa. The provider can add regions without notice.

For Global deployment types, the service can process prompts and responses in any geography where the model is deployed.

as of 2026-08-12 Source: learn.microsoft.com

Data at rest

North America: Canada.

Data stored at rest remains in the designated Azure geography.

as of 2026-08-12 Source: learn.microsoft.com

Never processed in

Azure in China, a separate cloud operated by 21Vianet

Microsoft Azure operated by 21Vianet (Azure in China) is a physically separated instance of cloud services located in China.

as of 2024-08-05 Source: learn.microsoft.com

Last checked

2026-10-21

Archived policy evidence

Northern Inference keeps a timestamped, hash-verified archive of this provider's data-use policy, so the legal basis that applied to any request can be shown after the fact.

The latest capture of this policy, on 2026-10-18, did not verify as the cited document, so no current archived copy is shown.

Partner-hosted routes. Some providers list models that a partner actually serves, for example Fireworks-hosted models on Microsoft Foundry. On such a route the provider shares your request with the partner, and the provider's own data-handling commitments do not govern the partner's processing. Northern Inference names the partner as the inference host on the route's page and in the request's custody record. Each dimension is stated only once the route's configuration (API mode, logging opt-ins, request storage) has been verified, every claim is labelled as resting on a published policy or on a contract, and such a route is never presented as zero data retention or no-training on the provider's word alone. Our Privacy Policy states this practice as its serving chain disclosure, and each such route's page lists the serving chain in order and the documents relied on.

Claims are sourced from provider documentation and dated. A "Not yet verified for this route" row means Northern Inference has not yet published a claim for that dimension. Postures are updated when provider documentation changes.

See the data-use posture explainer on the models page, or our Privacy Policy for the full data-use policy. Generated 2026-10-21.